Sybase数据库技术，数据库恢复专家

Only 5.0 Gold desplays appropriate behaviour. Broken in 5.02 and up, 6.0 and 6.01

----6.0 regression connecting to MS SQL Server. Works in all versions of 5.0, but broken in 6.0 and 6.01

SetTransPool() is not reusing opened connections until the maximum amount of connections allowed has been reached. Then opened connections are destroyed and reopened again

Solution:
This is fixed in Adaptive Server 11.5.1 ESD #4

Bug exists in 6.0.00 but is fixed in 6.0.01 after build 381.

SQLDBCode: 6

SQLErrText: CT_FETCH: Network Packet Layer: Internal net library error: Net-Library operation terminated due to disconnect.

A large menu on the MDI Frame contains many menu options that are visible and invisible. In the example some menu options are made visible at program execution time based on the number of menus requested from the user. The menu options are made visible using a FOR NEXT loop. When the count of options made visible is fairly large(seems to happen in test case between 66-68), the displayed frame menu is corrupted, that is, the menu does not display all the normally visible options and/or options made visible at execution time. In the actual application , when this problem occurs, the menu displayed is usually inoperative. By that, I mean that if you click on a menu option, the clicked event programming is not executed. This is not included in the sample as once the menu is painted correctly everything else works fine.

This is really the way it works for now. It's no different than the requirement in Powerscript.

decimal {2} id_decimal

id_decimal = 12.34 // not 12,34

dw_1.retrieve(id_decimal)

This area is not an editmask but a raw field so data like "12,34" when you bring it inside, becomes '12'. It's like doing:

decimal{2} id_decimal

id_decimal = decimal(sle_1.text) // doesn't understand "," so it truncates to "12"

Since this is a retrieval argument at the top level of a composite, this truncated value gets passed down to whatever nest needs it as "12" which would obviously bring back the wrong rows in a where clause compared to "12.34"

We know it's difficult for non-US users to think in terms of entering data in US format. This is especially confusing for end users - not just developers.

Perhaps the enhancement should be to use an editmask for these types of dialogs that would conform to the Control Panel settings.

Ex: #,###,###,###.## (+/- 2 billion can be represented). This isn't perfect since the user could enter (or want to enter 13,34355)

Workaround and better overall:

For Infomaker users there is no recourse at this point. The dialogs are system dialogs.

For Powerbuilder developers, design a response window dialog with an editmask:

decimal{2} id_decimal

Open(w_retrieval_args)

// capture 12,34 in the datawindow editmask the user typed in which gets translated internally in //Powerscript to 12.34

dw_composite.retrieve(id_decimal)

// if there are nests, the id_decimal will get passed through correctly as '12.34' internally or '12,34' as the end user understands it.

In a product like InfoMaker 6.0 where no custom dialogs exist, this can be a problem.

Possible workaround: Uncheck AutoSizeHeight in DetailBand properties.

To Reproduce:

1. Create a user object with a public method.

2. Create another UO that inherits from the first.

3. Create the COM object from the second UO (child).

4. View the IDL file for that object and you will notice that under that the methods section will be empty.

To Reproduce:

1. Create a UO with an instant variable as follows:

CONSTANT integer foo = 100

2. Save the UO.

3. Create the COM object from the above UO. (dual interface)

4. In the IDL file you will notice that under the "properties for public instance variables" section there is a propget and a propput property for the variable.

FIXED IN 6.0.01. DLL not specified.

( length_of_object modulo 4000 ) < 55

Under Unicode, the formula for the GPF would be:

( length_of_object modulo 4000 ) < 110

GPF can also occur for Library Import regardless of source control.

DBParm="UTF8=1 , DBTextLimit=32000 , PacketSize=8096 " will get none of these parms processed correctly. Moreover, there is no warning of any sorts to indicate that something was wrong.

Not reproducible on win95 and i don't have that font.

1/27 Tried this using Helvetica 12 and Helvetica 10 using JM's Sparcstation 5.
Could not get it to core dump.

1/30 Sparcserver 1000, SUN 4, SOLARIS 2.51 patch generic_103640-05
This problem is intermittent -- sometimes it works ok

4/13 TO REPRODUCE BUG:
Reproduced On GLARPO - PB 6.0:
created a dw with two columns and two headers.
shrank the height of one of the columns, selected other column, clicked on Edit/Size Objects/Make all heights same as first selected -- no problem
Shrank the height of the header text on the right, lassoed both headers, clicked on Edit

signal fault in critical section
signal number: 10, signal code: (, fault address:0x5a, pc
dd35384, sp: 0xefffe430
ABORT: core dump
Bus Error

Solution:
Cannot reproduce under build 444 pb 6.5 on Solaris.

*NOTE: This is not just with an EditMask of any type. SLE to SLE has the same behavior.

Developer's comments:

A deliberate change was made in 6.0 to have all events posted by PowerBuilder to be processed by a separate queue from the Windows system queue. As a result of this change, PowerBuilder posted messages are processed before Windows posted messages. Since the WM_CHAR message is a posted message, it now runs after the ue_after_keydown message. It is the WM_CHAR message that actually puts the character into the control. In otherwords, ue_after_keydown is running before the character is put into the control. The user should be doing his processing after WM_CHAR processing--WM_KEYUP would be a good place.

Solution Per Release Engineering:

A) a good point, we should use the same dir struct (requires adding an OLBOOKS dir to the Online Book CD)

however!

B) we specifically NOTE and attempt to ENFORCE short file names for online book installs because the Dynatext engine does not work particularly well with long filenames.

System Error: Cannot write to device AUX. <cancel> <retry>

The window will open and this will only happen once per boot session. This happens opening a window containing a native tab control in the window painter, running the window in development or in a pcode executable.

Development: Fixed in Pfcext60.exe of 6.0.01

PER DEVELOPMENT:

User error. New PBEOL050 and OCC020.DLL for 5.0.04 must be run together. They get installed in \pwrs\sys32 directory. User had an OLD VERSION of OCC020 in his \winnt\system32 directory which gets loaded first. This old version does not have the cFREEblob function implemented.

In the documentation, under the Functions for DataWindowChild objects we state that you can print child datawindows up until 6.0.00. This was always incorrectly documented and never supported. For 6.0.01, Print() has been taken out as a method for DataWindowChild.

The Web.PB NS-API 2.0 interface pbns2050.dll and pbns260.dll will cause the Netscape Enterprise 2.01 web server to crash when a request is cancelled or refreshed before Web.PB has finished the original request.

Note: This happens only with Netscape Browser version 3.0, but it works with version 4.0.

Product Support Engineering:

March 31, 1998 - StephenD
6.0.00 - does not bring the correct tab forward.
6.0.01 build 390 - works fine. Fixed with other fixes to posted events and tabs

Two workarounds:
A)
Set the selectedtab properties in a timer event and after set timer to 0
Open event : timer (0.1)
Timer event : My_tab.selectedtab = 3 , timer (0)

B) Keep the scripts existing and add the following ...
Use an instance long varible ilcp=0
In the selectionchanging event :
if ilcp = 1 then
return -1
end if
ilcp++

10/07/98 Simon Besher.
Yet, another workaround:
In constructor event of the tab control
This.Post Function SelectTab(#)
or open event of the window
tab_1.Post Function SelectTab(#)

Ex:

1,234.56

1x,234.56 // cursor is at the 'x' for display purposes.

If you hit the delete key, it appears nothing happens. It always worked this way. If cursor is at 'x' above, it doesn't pull data from beyond the delimiter (to the right). The character you're trying to delete is the ',' the delimiter. Same behavior if you highlighted the ',' and tried to delete it.

*** 6.0.00 Regression: Originally reported in 444267, bug#44118

EditMask for datetime field, if you enter a time as 00:00:00 it will change to 12:00:00

- See issue 455165 which has the reverse behaviour in 5.0 - enter 12:00:00, the time will change to 00:00:00

Per Development:

This example has a datetime with a real date value and a null for the time. It is not truly NULL so we have to display it as we see it according to the mask.

In working across the mask, we find AM/PM and decide that 00:00:00 is actually 12:00:00 AM.

You can avoid this by removing the AM/PM from the mask, then we will display the 00:00:00

1. After SetfullState() using a graph to a graph datawindow, nothing shows up. If you then exit the window, it GPFs. By issuing a Filter() or Sort() right after the SetFullState(), this forces a repaint and it does show (workaround) and also it doesn't GPF when closing the window..

2. After SetFullState() to a crosstab to a crosstab, I get "out of memory" even in NT 4.0.

If you use a Tabular to GetFullState() and Setfullstate() to a crosstab already assigned to the destination control (so it uses that one instead), I don't get an 'out of memory' and the crosstab seems to display fine.

3. Nested datawindows GPF on SetFullState()

4. Crosstab datawindows give the error "load of nested report failed" when Setfullstate() is executed.

GPFs in 6.0.00 Gold but not in 6.0.01 build 347.

FBASHAW: Verified fixed in 6.0.01 Build 419.

Also need these statements in the following order (they were reversed):

session.Connection = connection
session.Connect(0)

To Reproduce:

1. Create a new COM project.

2. Select the UO that you want to use.

3. Set the properties in the properties dialog.

4. Build the component.

5. Close the Project painter.

6. Open the UO. You will get the following error msg:

"UserObject xxx, one of its ancestors or descendants, or a control inherited from it, is already open in another User Object or Window Painter and cannot be modified."

There are no other painters open. Also, the component does not have to successfully build, it seems as long as the UO is accessed you cannot open it again without restarting PB.

PB050 has caused a 10H exception in module PBBGR050.DLL.

Print Preview works OK.

Main DataWwindow sharing data with two or more other datawindows. If SetDetailHeight() is called to collaps the rows on the secondary datawindows and datawindows are Grouped By (main and secondary,) the arrow buttons on the scrollbars of the secondary DataWindow's are not functioning.

Workaround: If Bring To Front is chosen for the rectangle the gpf does not occur.

(In PFC Examples, Datawindow examples, Row Selection example - using "Single Select", this manifests itself by not allowing the highlight of a new row - ie it takes 2 clicks to highlight a row.)

decimal{2} l_dec

l_dec += dw_1.getitemnumber(1, "decimalcol")

l_dec += dw_1.getitemnumber(2, "decimalcol")

or

l_dec += dw_1.getitemnumber(1, "numbercol")

l_dec += dw_1.getitemnumber(2, "numbercol")

GPFs in 5.0.04 and 6.0.00. The workaround:

decimal{2} l_dec, l_temporary

l_temporary = dw_1.getitemnumber(1,"numbercol")

l_dec += l_temporary

Developer: PB5 has a hard coded limit to number of characters which used to be 16 and is now 30. PB6 does not have a problem.

Escape sequences for printing not working in machine code executable. (It does work when run from development environment and compiled pcode)

Developer: User must rebuild their exes with the new PBCMP60.DLL

- the leak does not occur when doing either function separately.

Should be

dwo.primary[row]

not

dwo.data[row]

The problem was fixed for Win95 and Win NT 4.0, but is still broken on NT 3.51 due to operating system bug (per developer).

From Development (3/3/98)

This is only a problem under NT 3.51 if there is no initial selection and the item is the first one in the list (Typing 'A' works as expected). It appears to be an OS bug as it happens in the Non DW DDLB as well. This continues to happen even when we disable all of our window's subclass processing, which further points towards an OS bug. A work around is to have an initial selection and the problem does not occur.

The change from 5.0 to 6.0 is probably due to the fact we mark our 6.0 executables differently so that everything work properly under WIN95 and NT4.0. It looks like this may be causing NT3.51 to behave differently.

This CR is a Dupe of CR 156272 which is a Microsoft issue:

This is an operating system problem. It has been reported to Microsoft. The reference number for the issue is: SR X99041360189

Enabling the WEBPB tracing on the transport object will cause the dpb server to crash when the argument string length exceeds approx 400 characters. The trace console window displays all of the CGI environmental variables including the long argument. It crashes when it is writing the long string to the console. The DPB server program only crashes if the transport.trace webpb=1 logging is enabled.

The only way to prevent this is to disable the WEBPB tracing.

Two error messages appear:

1) pb60.exe - Application Error, "The instruction at "0x36353433" referenced memory at "0x36353433". The memory could not be read.

2) pbcgi60.exe - Application error, the instruction at 0x100034d5 referenced memory at 0x00000000. The memory could not be read.

Again, this works fine if classes are local and proxies are not used. In this case, there are some proxies leftover from 5.0. (isn't this supposed to be upwardly compatible?)

Development: Developers should be either using one way or the other. The 5.0 way was with a proxy and "Create using". The 6.0 way is using CreateInstance() for server or "Create Using..." for local with NO proxies. Mixing the two will cause this problem.

Pressing the the backspace key in a SLE control is sending the keystroke to the browser instead of the SLE.

The problem appears to have been corrected by itself in newer builds of the pbrx60.ocx. I checked with the developer and he wasn't aware of this problem.

ldw_dw and la_dwarray[1] were populated with the dw_1 reference.

Easy workaround is to right click control and hit duplicate and it works correctly and does not gpf. If you then try to Edit = > Copy = > Edit =>Paste the copy it works fine. PB 6.0 and 6.0.01 work correctly and do not exhibit the problem.

Dot notation assignments to rows in filter buffer cause a GPF.

Problem: Using SYC connection, a dynamic execution of a stored procedure causes DB error -137 - "Must declare variable '@return_code'" when the current user is not the owner of the procedure.

Workaround:
-----------------
Instead of using the same object for both reports, save the object with anoter name (Save As) and use it for additional report. It will increase the footprint of the app, but will save PSR without reflection the changes on the second report.

Developer: The fact that this worked in 5.0 was due to an unrelated bug. It cannot be fixed in a maintenance release. Simple workaround is to assign something initially and then destroy it--IsValid will work correctly after the first time through.

###,##0.00;-###,##0.00;0.00;DIV0

but it always shows 0.00 when previewed instead of DIV0 as expected. Rebuilding the computed field somehow fixes it but there is no guarantee what you are seeing is correct.

Does not GPF when autosizeheight is turned off for the detail band. It must be related to trying to paint some nested datawindows and not others and trying to slide all the information up at the same time.

"The instruction at 0x11111d87 referenced memory at 0x01111000. The memory could not be read"

To Reproduce:

1. Create a COM Project.

2. Open the 'Select Objects' Dialog.

3. Select an object and then click on the 'Apply' button.

Nothing will happen, it should behave like the 'OK' button minus closing down the dialog.

In the Datawindow Painter in the SQL painter, when converting from graphical to Syntax, if there is an error (column name missing or changed), the SQL statement does not line wrap making it some what unmanageable.

This happens for simple ' ' or 9 as computed columns (where the value is the same during retrieval - or when you're using backend functions.

This happens in 5.0.02-5.0.04.75 thru 6.0.01 build 375 so it's not a regression but very bad painter behavior during the re-describe step.

If a HTML page has a form that doesn't have any input or hidden fields, the POST method will hang when calling Web.PB. The browser will just wait indefinitely. The problem occurs with the PBCGI60.exe, PBNS360.dll and PBNS260.dll. The ISAPI pbisa60.dll works correctly.

The work around is to add a dummy hidden field to the html form.

The migration actually changes the x,y,width and heigth of the object. (the example shows PBUs)
5.0 example SLE
x=1130,y=353,w=577,h=97
6.0 after migration
x=1129,y=352,w=576,h=96

Workaround: Edit all the SLE's and make them slightly bigger.

Application Error:

"The instruction at 0x1113bad4 referenced memory at 0x00000000. The memory could not be read."

NOTE: We have a few Halt Close issues. They are not all the same as user can code just about anything in the close event of the application.

Problem: datastore.object.data.<buffer> dot notation access in the SelectionChanging() event of a tabcontrol prevents Return 1 from working.

Page 1 of 2

Page 2 of 2

Page 2 of 2 (on the overflow page)

Now, in 6.0.01 Build 365- 375 you get

Page 1 of 2

Page 3 of 2

Page 4 of 2

This is a "change in behavior" of sorts but what the customer is doing will always mess up the pagecount() since his data for one detail row extends beyond one page - the current limitation of the datawindow.

I guess the problem is we handled it consistently up through build 360.

*** Workaround

Place an empty text object on the datawindow instead of a calculated field

In the window with the datawindow control, create an instance variable of type LONG that will contain the maximum number of pages to be printed.

In the PrintStart event of the datawindow control, assign the instance variable the PageMax argument - this represents the last page being printed

In the PrintPage event assign your page count to the text attribute by using PageNumber argument and instance variable.

i.e.

INSTANCE VARIABLE

long ll_lastpage

PRINTSTART event

ll_lastpage = pagemax // the pagemax is an argument to the event

PRINTPAGE event

// page_count is the name of the text field in the datawindow

// pagenumber is the current page being printed passed as an arugment to dw control

dw_1.object.page_count.text="Page: "+string(pagenumber)+" of "+string(ll_lastpage)

This works in 6.0.01 build 375

This issue occurs when you follow these steps:

1. Create a uo wich contains for example a dw. In the dw constructor create a popup menu. In the dw destructor, destroy the popup menu.

In the rbuttondown for this dw, call the function popmenu(xpos, ypos)

2. In the menu script uses for the popup menu, write something like :

MessageBox('Parentwindow', ParentWindow.title)

ParentWindow.TriggerEvent("ue_event")

Execute and/or debug the application, you can see that the parentwindow points to the uo instead the window.

Workaround: As stated, if you get a REAL datatype in row/column specifications against a decimal(n,0) column, you must export and change it to decimal(0) in the datawindow.

Problem: Unable to right-click and drag to select multiple items in a listview.

Problem: Unable to call a procedure in an Oracle package that is declared as an RPCFUNC using OR8 native driver.

The "Compare Differences" Menu item is enabled in the PopUp menu when you rightclick a registered object. If you go to the Source Menu, the "Compare Differences" menu item is disabled.

Since this ability did not exist in 5.0 and no new enhancements are being made to the native interfaces, it should be disabled. It should be disabled in the popup menu as well.

Only the SCC interface has the "Compare Differences" functionality enabled.

Application Error:

"The instruction at 0x110b85c8 referenced memory at 0x61616179. The memory could not be read"

Workaround:

Use a shorter path.

Select error: SQL0104N An unexpected token " " was found following "". Expected tokens may include: "+ - .... SQLSTATE=42601

The GetURL and PostURL methods do not support connections to HTTPS / SSL web sites. When accessing a secure web page ''General Error'' will occur.

The work around will be to use an ActiveX control that supports SSL.

Occurs with the Microsoft SQL Server driver (PBMSS60.DLL)

has bogus characters at the end if the computed field contains the (3-argument) mid-function, the 3rd argument is larger as the 2nd, and the 1st argument is large.

Workaround until this is fixed:

Add the following code to the constructor event of the edit mask:

this.SelectText(1,11)

this.Replacetext("00:00:00 AM")

this.SelectText(1,0)

This will initialize the value to what it was in PB 5. However, if the user selects the entire field and clicks <delete>, it will revert back to the PM value.

.Category and Database:
Sybase SQL Server 11.0.3

.O.S:
HP-UX 10.20 (PHSS_10053, PHSS_10113, PHSS_10565)

.Sybase Products:
PowerBuilder Enterprise 6.00 for HP-UX

.DB Interface:
libpbsyd60.sl
OpenClient 11.1.0

.Type of network:
Not Applicable

.Does this problem occur in development or/and at runtime environment ?
Development

.If there is other windows software involved ?
No

.Description:
When trying to print a datawindow with a compute field ''hidden'' outside the print area causes CPU usage overhead and nothing is actually printed.
The compute fields mentioned above handle bitmaps, however are others like that on the print area and it doesn't cause any problem at all.

DETAILS:

Using a spin control editmask on a date column in a datawindow, if you spin to March without entering anything else, you'll get a different value in the date depending upon the control panel setting and upon the editmask. The Control Panel influence makes more sense and is probably by design, but the other doesn't make sense:

Scenario 1:

CONTROL PANEL SETTING FOR DATE: mm/dd/yy:

Editmask -------------spin to March-------> result displayed -----------> date

[date] 03/01/00 March 01, 2000

mm/dd/yyyy 03/16/1998 March 16, 1998

mmm/dd/yyyy /00/0000 NULL

The difference in the first two makes sense since "00" is a valid year, 2000, but 0000 is not. We seem to default to today's date. However, if you spin to March on the mmm/dd/yyyy mask, why doesn't it work the same as spinning to "03" for the month using the mm/dd/yyyy editmask? Why don't we either default to today's date for both, or for neither ?

Scenario 2:

CONTROL PANEL SETTING FOR DATE: mm/dd/yyyy:

Editmask -------------spin to March-------> result displayed -----------> date

[date] 03/16/1998 March 16, 1998

mm/dd/yyyy 03/16/1998 March 16, 1998

mmm/dd/yyyy /00/0000 NULL

Again, why do we default to today sometimes, but not others ?

Scenario 3:

CONTROL PANEL SETTING FOR DATE: mm/dd/yy:

Editmask -------------TYPE March-------> result displayed -----------> date

[date] 03/01/00 March 01, 2000

mm/dd/yyyy 00/00/0000 NULL

mmm/dd/yyyy /00/0000 NULL

Scenario 4:

CONTROL PANEL SETTING FOR DATE: mm/dd/yyyy:

Editmask ------------- TYPE March-------> result displayed -----------> date

[date] 00/00/0000 NULL

mm/dd/yyyy 00/00/0000 NULL

mmm/dd/yyyy /00/0000 NULL

Scenarios 3 and 4 are probably by design, and they make more sense because of the system date settings.

string dbms = Left(string_arg_passed_to_funct, 3)

return 0

GPF goes away if you do assignment rather than initialization

string dbms

dbms = Left(string_arg_passed_to_funct, 3)

This appears to be the same problem as in issue 496749/bug 49432.

When a datawindow is saved as a PSR on HP and then displayed on a Windows system in the DataWindow Plugin, the plugin will get an out of memory error. This only seems to happen if the Datawindow has a retrieval argument.

Also, an Application Error will result:

"The instruction at 0x112066b2 referenced memory at 0x00000054. The memory could not be read"

Solution:
Per document Id 47860 "PowerBuilder 6.5: 16-Bit Applications on 32-bit Platform" (available on the Sybase website) deployment of 16 bit executables on 32 bit platforms is no longer supported. This will not be fixed.

The EditMask for a decimal format "##,###.00" behaviour is inconsistent when tabbing from column to column. The editmask also allows the user to enter non-numeric data (alpha characters).

This problem appear to only effect Unix.

Does not happen in 6.0

Per Development:
Not to be fixed.

Text file created only contains a PageBreak Code.

Works Properly under WinNT.

Workaround:

Use a combination of the Print() and PrintText() functions, see online help or page 1092 of the Powerscript Reference vol2 (ver. 6)

When expanding a treeview item the ItemExpanding and ItemExpanded events fire twice in the following order: ItemExpanding, ItemExpanded, ItemExpanding, ItemExpanded.

The double triggering typically appears only on the first item but on occation you will notice it on other items. This typically happens when you click once on the icon for item "A" and then doubleclick on item "B" to have it expand.

Tested PB 6.0 GA, 6.0.0.385 and 5.0.04.75 and all show the same behaviour

UPDATE: The problem seems to be that InsertItemLast does not like being passed a treeviewitem variable which was used previously to reference the parent rather than the child item. I would say this is not a bug except that the problem only occurs for the first item. It is easy to workaround: either use two different treeviewitem variables in the code, one for all references to the parent, or "null out" the treeviewitem variable you are using by setting it equal to a treeviewitem variable which has not yet been used before you call InsertItemLast. In either case, you have to declare two treeviewitem variables.

This may be a bug, or it may be that the documentation for InsertItemLast needs to be expanded to include more information about how the treeviewitem argument will be used by the function on input, and exactly what initialization is expected.

CODE THAT CAUSES PROBLEM BEHAVIOR:

// **itempopulate event** //

treeviewitem ltvi

int i

string ls_item, str

this.GetItem(handle, ltvi)

ls_item = ltvi.label

FORr i = 1 to 10

ltvi.label = "Item " + string(i)

ltvi.children = false

ltvi.pictureindex = 2

ltvi.selectedpictureindex = 2

tv_1.InsertItemLast(handle, ltvi)

next

CODE CHANGE OPTION ONE:

// *itempopulate event* * * use another treeviewitem for the populate code **//

treeviewitem ltvi, ltvi_child

int i

string ls_item, str

this.GetItem(handle, ltvi)

ls_item = ltvi.label

for i = 1 to 10

ltvi_child.label = "Item " + string(i)

ltvi_child.children = false

ltvi_child.pictureindex = 2

ltvi_child.selectedpictureindex = 2

tv_1.InsertItemLast(handle, ltvi_child)

next

CODE CHANGE OPTION TWO:

// **itempopulate event** //

treeviewitem ltvi, ltvi2

int i

string ls_item, str

this.GetItem(handle, ltvi)

ls_item = ltvi.label

//before the FOR loop, "reset" the treeviewitem variable to "null" by

// using the second treeviewitem variable, which is never otherwise used.

ltvi = ltvi2

FORr i = 1 to 10

ltvi.label = "Item " + string(i)

ltvi.children = false

ltvi.pictureindex = 2

ltvi.selectedpictureindex = 2

tv_1.InsertItemLast(handle, ltvi)

next

This works fine in PowerBuilder 6.0 and an easy workaround is available. This will not be fixed in Powerbuilder version 5.0.

The example shows a resource drain due to a tight loop the user is running for stress testing..

A correct day name will be displayed even if the year is NOT a leap year.

Application Error:

"The instruction at 0x77e72f61 referenced memory at 0x00000000 the memory could not be read"

B) Computed fields that contain a tab character (~t) display only a white square box - not a tab

C) Computed fields that contain a tab character ( char(9) ) display only a white square box - not a tab

Workaround: Any user object function taking readonly or reference arguments will be created incorrectly in the html output. As you select the function in the wizard, if it's readonly or by reference, make sure you edit the output.htm before you use it.

The instruction at "0x11313858" referenced memory at "0x00000002". The memory could not be "read" error.
The instruction at "0x11313829" referenced memory at "0x025110be". The memory could not be "read" error.

This often is found with databases containing large text columns like SYB MS SQL Server 4.x, Microsoft SQL Server 6.5, Oracle, etc but this can be recreated simply even with an external datawindow with a char(518) column populated with data and having a script call "GenerateHTMLForm()".

There seems to be a 5:1 ratio of virtual memory usage

10 sheets opened in an example:
5.0.04 - (3 MB)
6.0.01 - (15 MB!!!!!!)

5.0.04 used much less memory.

Development: This leak was associated with large heap usage with the datawindow and is fixed in the first EBF to 6.5.00 in build 454.

The following DLLs are left resident in memory:
PBBGR60W.DLL
PBDWE60W.DLL
PBOUI60W.DLL
PBVBX60W.DLL
PBVM60W.DLL

Simple workaround to unload the dll's left in memory:

Declare as Global External Functions:

SUBROUTINE FreeLibrary(uint libhandle) LIBRARY "Krnl386.exe"
Function int GetModuleHandle(string modname) Library "Krnl386.exe"

Script for application close event:

uint l_handle
l_handle = GetModuleHandle("pbdwe60w.dll") // This can be any dll or library that is loaded into memory. Do not need qualified path if on DOS path
FreeLibrary(l_handle)

Workaround/Resolution: Do not use SetRedraw(false) in the openevent of the frame.

Per Development:
This is not a bug, but expected behavior.

The Transaction Pooling functionality should reuse connections in the pool that matches previously released connections with the same transaction properties (i.e. DMBS, Database, etc.). The fact is that the transaction manager will use all connections up to the maximum and then close any pooled connection before re-establishing the identical connection.

During the testing I also noticed that occasionally connections would be loss. The connection would remain at the database even after PB was closed. I have been unable to consistently reproduce this problem.

1. create an NVO from an oleobject

2. add externalexception script to it

3. assign an oleobject returned from a method call to it

4. the scripts, udo functions, instance variables, etc are no longer addressable.

Cannot reproduce 6.0 build 213, 5.0.04.48

string ls_args[], ls_types[], ls_temp[]

or

string ls_args[], ls_types[], ls_temp

ls_temp = ls_temp + ls_args[] + " (" + ls_types[] + ") "om

In the table painter, when connected to a MS-SQL Server, the definition of a decimal column allows the user to specify how many decimal places this column will support. If you actually try to specify a non-zero value for this field (like 2 decimals places for a dollar amount) you will receive the error "the maximum scale for decimal columns is 0".

This error appears to be specific to MS-SQL Server 6.0 and to the Table Painter as Sybase System 10 handles this correctly - as does MS-SQL Server 6.0 when executing the same DDL from the DBA Notepad. Also, existing tables show up with non-zero decimal places in the table painter.

Workaround: Write out the query first in the Datawindow painter syntax mode, and then converts to graphics, the table representations will appear.

The fix for PBODB050.ini file in the SQL Anywhere section - Add:

TableListType='GLOBAL TEMPORARY'

If you do not have permission to make certain changes to a table, PB should tell you this. In 4.0 it did. But in 5.0, if just ignores the command and gives no message.

PER DEVELOPMENT:

Basic change of behavior was introduced in PBSYB050 to help with ptrack #282918. Development management wants us to go back to our historic PBSYB behavior.

In releases 1 thru 4, Print statements from stored procedures were treated as fatal errors. Only exception was an explicit list of success messages from sybase system procs.

Customers often want to use PRINT statements for debugging stored procedures, so in PBSYB050 and PBSYC050, my intent was to ignore PRINT statements except an explicit list of fatal errors returned by sybase system stored procedures like sp_rename, sp_dropkey, etc. These errors from sp_dropkey were missing from the new list.

Development management believe that to solve this problem, we should revert back to the historic behavior of PBSYB rather than add these messages to the "fatal error" list. Thus, when this change is applied to PBSYB050 maintenance, our behavior wil be the same as in release 1 thru 4 described above.

PBSYC050, on the other hand, has a much shorter history. So it is more acceptable to innovate in PBSYC. PBSYC050 will continue to ignore PRINT statements from user-defined stored procedures and ignore success messages from sybase system procedures. Failure messages from sybase system stored procs are treated as fatal errors. No changes are needed to PBSYC050 for this ptrack issue. The bug was strictly in PBSYB.

In DB painter, he can see all his tables and views, but when he double-clicks on the view (looks at the extended definition),

In 4.0.0x he gets message box with a stopsign, but he can't see any text inside.

In 5.0 he gets Dr Watson error and gets booted out of PB. page fault in pbo72050.dll. Also happens with o71 interface..

This happens when you are not logged in as the creator of the view.

Developer: The example coded causes the infinite loop hence the ''infinite looping'' is expected behavior.
Should use ''ExceptionIgnore!'' instead

not bringing up login dialog. 'Database Error: Database connect failed, Application cannot be run'.

SQLCODE=-1, SQLERRTEXT="SQLSTATE = IM008 [Microsoft][ODBC SQL Server Driver] Dialog Failed

string ls_mystring = global_array[1]

in the open event of the application, Powerbuilder fails in PBRTE050.DLL and changes to the script are lost.If you declare a global unbounded array of string, and then reference that global array by as follows:

string ls_mystring = global_array[1]

in the open event of the application, Powerbuilder fails in PBRTE050.DLL and changes to the script are lost.

WORKAROUND:Changing the reference to:

string ls_mystring

ls_mystring = global_array[1]

worked just fine.

return noconnectprivilege!

but if this happens, you cannot save any user objects when you're returned into powerbuilder. You have to exit Powerbuilder and re-run.

Error message: request failed possible causes:

1. library with complete path\pbl not found or

2. library can be used in read mode only

Workaround: Use Getitemnumber() first and check if isnull(). If true you're done. If it is

not null, do a getitemdecimal().

PER DEVELOPMENT:

Like Open, if you call OpenSheetWithParm using a variable that points to a valid window instance, the window is activated instead of opening a new window. Use another variable or use SetNull on the variable between calls.

Cannot reproduce - 5.0.04, 6.0

However, the same 'c' prefix problem happens with NORMAL joins using synonyms - this bug should be addressed. Also, and more importantly, the syntax generated doesn't have table qualifiers.

NOTE: For using SyntaxfromSQL() in runtime - Not a bug. Must set USERID to the OWNER of the table (not necessarily the same as LOGID in most cases).

This is a nuisance since you must edit all the datawindow column names but there is a workaround.

Cannot reproduce 5.0.03, 6.0 beta.

WORKAROUND: Use 32 bit Powerbuilder for the migration. Customer has an EXTREMELY large number of objects and has probably hit an internal limit in the 16 bit world. This application will also have trouble compiling into an executable (machine or pcode) - aborting in the same 'inspection' phase using the 16 bit development environment. Should use 32 bit throughout on applications this large.

For those that need 16 bit PBDs (99%) rather than machine code, there is a way do create a 16 bit exe stub using the 16 bit development environment. PBDs are transparent and can run on 16 or 32 bit platforms so the process to build a 16 bit PBD executable with large applications that won't compile is:

1. Build the pbds using 32 bit PB pcode.

2. Create a seperate directory for 16 bit since you'll need to modify some of the pbls,etc

3. Make sure your library list is correct but delete all objects not referenced in the application object. The application object might have global variables, global functions, the MDI frame, etc. This process will get your referenced object count down.

4. Build the 16 bit PBD exe. This should take seconds since you might only have 20-50 objects referenced in the application object. This stub can be used over and over unless you add or delete pbls from your library list.

You can then use the 16 bit stub PBD exe with the PBDs you built in 32 bit. At runtime, all the other references should be resolved fine.

NOTE: Not all objects in all pbls are counted during the inspection phase. Objects referenced in strings and any objects they might call aren't included during the inspection phase so the 'size' of an application isn't the true indicator but the number of 'referenced' objects - objects 'seen' by the compiler is.

/* w_frame will be counted below */

open(w_frame)

/* w_frame, its instance variables, calls to other objects, etc will not be counted and this could make a significant difference. Other references using open...() syntax 2 also aren't counted */

window wi_frame

open(wi_frame, "w_frame")

Applications this size won't be a problem in Powerbuilder 6.0 since only a 32 bit development environment will be offered with the functionality of both 32 and 16 bit environments including 16 bit pcode generation.

NOTE: This limitation will not be fixed in 5.0 16 bit.

For 5.0.04 32 bit development, 6.0 functionality exists to now build a 16 bit pcode executable using the 32 bit environment.

WORKAROUND: There are a few ways to work around this. One would be to create two datawindow objects with identical result sets and modify the display of one to show the columns in the new position. Then use both dataobjects on the window, do a sharedata to share the result set, and set the visible properties so that the desired display is the one shown. Another possiblity is to show the columns in the datawindow more than once in different positions, and conditionally set the visible property on the column. The best workaround depends upon the requirements of the particular application.

PER DEVELOPMENT:

Solution:User error: a number of things jump out with this ptrack: the fields not showing are because the last field that's been moved to the first position is a wide field and, since it's painted last, paints over those 3 fields (i.e., the logic in the f_move_columns function is adding the wrong amount to the new x position). Also, there's a computed field that has the same name as a column (this is certainly not possible to do these days through the painter - maybe it was back then -or- maybe they did something funky through syntax), this could account for the occasional gpfs.

This only occurs in the executable. Everything works fine in development.

The same code works using Powersoft Demo DB V5 with ODBC driver for both machine code and p-code exe's.

PB 4.0.

PER DEVELOPMENT:

There is currently no way to determine if a stored procedure is a system procedure or a normal procedure via ODBC. ODBC provides functions to identity table types but not stored procedure types.

After a combination of mouse clicks buttons become disabled in PowerBuilder. They are unusable until you restart PowerBuilder. I have duplicated this with 100% accuracy on my computer, another PC in the lab this will only occur 50% of the time. This has been an ongoing problem with no way to duplicate it until now, so I am submitting this PTRack even with the 50% result on the other machine.

To Duplicate:

1)In PowerBuilder switch to the demo .pbl that comes with PowerBuilder. Example50 is the name of the application. If this .pbl is not available another large .pbl works, the larger .pbl produce more consistent results.

2)Quit PowerBuilder. And ShutDown and Restart windows95.

3)When you get back in Windows 95 go into PowerBuilder.

4)Click on the Application painter.

5)Click on the Window painter and open any of the windows.

6)Click on the Menu painter and open any of the menus.

7)Then click on the DataWindow painter and open any of the DataWindows.

8)Click the preview mode and then exit out of the preview mode.

If by this point you have not seen some of the button along the top of the screen turn grey and become disabled repeat steps 7 and 8 again.

Same as bug 30460

FROM DEVELOPMENT: there were two problems demonstrated in the sample PBL.

The first bug demonstrated is not a bug. The customer has a function on a user object that takes an unbounded array by reference. In script, the customer is passing in a bounded array. This is just not allowed. It doesn't work ever. The types must match exactly when passed by reference. They include a "comparable" example with arrays of strings, but in that case they pass in an unbounded rather than a static array--which is demonstrably not the same thing.

The second bug they demonstrated, dynamic call of an event with structures passed by reference works correctly as of the latest build (5.03.26, fixed in pbrte050.dll).

Calling the same function statically does not cause an error.

Same thing with user defined events called dynamically get error PowerBuilder Application Execution Error (R0003). The error message states "Application Terminated. Error: Array boundary exceeded..." Calling the same function statically does not cause an error.

Note:

Having an array of strings with the event/function either being called dynamic or statically does not cause gpf.

===============================

The customer has sent a sample app. that reproduces this issue. I have tried to put a boundary on the array but still got the same error.

Table with key that is a double and is used as an autoincrement. Created a datawindow and when a new row is inserted, the autoincrement value does not show unless a re-retrieve is done.

He exported the syntax from the datawindow and compared it to one that the key was an integer autoincrement and found that the syntax was different; no identity=yes for double autoincrement.

WORKAROUND:
Customer added identity=yes to the export and reimported the datawindow or change the datatype to integer which will limit the number of records that the table can hold.

Development: Added support for doubles as autoincrement defaults. Fixed in pbodb60.dll and pbodb050.dll. Note: To test, you must create a new datawindow instead of using the existing datawindow. This is because the identity attribute is set as the dw is built.

Library: c:\path\retrieve.pbl

Object: uo_retrieve

Forward Prototype Declarations

(0002): Error C0122: Function argument differs from ancestor only by pass by reference or value for function retrieve

Original Bug Definition: User does not receive valid values for dates when using rpc's and the net gateway. Using date format other than ISO.

revised by lindak 11/14/97

NOTE: Retested in 5.0.04.10 Win 3.11. The fix for bug #41071 for OpenTab/CloseTab memory leak in 16 bit seems to have resolved this one as well.

Error 999: Invalid stored procedure parameter mode of 'out' 'begin' please change to 'in'

It appears that Oracle is returning a complete description of the result set (see describe below). Our drivers are expecting only one line back, the ref cursor.

This works with Oracle 7.2.2.4.0, but not 7.2.3.0.0 and 7.3

Per Development, this in not a bug: Can't use the same shortcut key twice in the same menu.

Workaround: Use an invisible datawindow control. This has been fixed in 6.0 only.

When previewing a dw that has retrieve as needed gets "select error sql -932 error on network connection gethostbyname(wsano_data) system call failed. has occurred". Customer is using 5.01 wh1 of inet.

Only when "retrieve as needed" on, you get a

"Select Error:" messagebox. The database trace shows:

SERVER=informixserver (2073 MilliSeconds)

Error -932 (rc -1)

Retrieve as needed is opening up a second connection (cursor). 183898 mentions to increase the # sessions on server to 3 (for a particular process) and the problem went away.

tab order. Populate the treeview on a post open event.

If the first keystroke the user clicks is the right arrow button on the

keyboard ->, they will get a invalid page fault pbshr050.dll @ 014f:11169c84 (In 5.0.02 on Win95, address was 0137:11169F84)

Workarounds

1) In code, expand the first item and select the first item on the treeview control. The

first item being the first item that was inserted onto the treeview control.

This gets rid of the GPF's.

2) Another workaround was to put the population code in the open event, rather than posted to custom event.

DUPLICATE OF BUG 37085.

Workaround: Post the halt(close) after the dberror event

The problem occurs only on 16 bit versions (4.0.05, 4.0.06 and 5.0.01) and only for strings. It seems there is a SQLBindCol missing (or it is a ODBC.DLL problem).

"ct_bind: user api layer: external error an illegal value of 2 given for parameter item."

Workaround: Use staticbind=0. This way, it behaves like 4.0 did where it would describe before retrieve to match up columns.

Development: This should be documented that the user has to use Staticbind=0 with these types of modifications.

This was fixed with a version of PBROI050.DLL that shipped after 5.0.02 (11/5/96)

TWO PART SOLUTION: The bug is fixed in 5.0.03. Also, you cannot connect to a database in the ConnectionBegin event unless you set the Transport Dedicated by coding:

transport.options="Dedicated=1"

This is not documented for 5.0, but will be documented for 6.0. See also bug number 42042. Using this option also solves that problem.

PER DEVELOPMENT:

User error: Cannot select char columns into PB datetime host vars (columns are misnamed). Change l_datetime & l_smalldtm to string (l_bit to double) to correspond vars w/cols.

This was a long-standing deficiency in the product. Unfortunately, we could not fix the NUP case at this time.

double d_a[1] integer i_a[1] SetNull( d_a[1]) i_a = d_a // this will GPF

integer i_a[1]

integer i2_a[1]

SetNull( i_a[1])

boolean b

b = IsNull( i_a[1]) // b will be true

i2_a = i_a

b = IsNull (i2_a[1]) // b will be false

FIX is for GPF. Preservation of NULLs submitted as a new bug: 44501

The problem becomes, if you select, say "Landscape" orientation from one of the "duplicated" entries, it's an invalid entry and you don't get the orientation you'd expect - most often Print.Orientation, for instance, gets set to 0 (default). You can go back and change it in the painter but you do have to be aware of the bug.

PB050 caused a stack fault in module PBDWE050.dll @ 001e:000e

Development: Intensive numeric computations causes this. The compiler option to optimize is causing this problem. Took out optimization for 16 bit and it no longer has the problem. DLL affected: PBDWE050.DLL 16 bit.

A "Select Search Path" dialog appears "Unable to locate the specified file in any of the specified search paths. Discover while attempting to display help file "IMHLP60.HLP" (or PBHLP60.HLP).

You can close this screen and online help works but when you close IM/PB and try again, the problem is back, even when you 'set' the option.
The problem is PB/IM is trying to write a small temporary file (1.2) to the network directory (why you need write access) which should be redirected to the users local \temp directory but it doesn't.

This is also true for the PB and IM demo db files - a file named 1.2 is also created then immediately deleted when the databases are launched..

Workaround: Give the user(s) write access to where the help files are.

Development: Definitely broken in 5.0.02 and 5.0.03. Fixed in 5.0.04 and 6.0 by other changes to the datawindow.

select emp_fname

from tsdata..employee table1,pubs2..department table2

where table1.dept_id = table2.dept_id

and table1.emp_id = 105

Notice the two databases "tsdata" and "pubs2". This SQL works in the DBA Administrator and in the pipeline painter (you have to use Convert to Syntax to type the SQL in, the SQL painter will not paint this graphically.). When you try to save the pipeline object which contains this unconventional SQL statement, you get a GPF in PBSHR050.DLL.

Even a comment in datechanged event causes a GPF when trying to open the tab pages since during the development environment construction of the tabpage, it seems to somehow be firing the datechanged event.

Does not GPF in 5.03.29, 5.04.10 or in Panther.

If a userobject function returns a blob and has an argument of type integer and the corresponding value that is submitted to web.pb is non-numeric, the following error is returned by WebSite:

500 Server Error

The server encountered an internal error or misconfiguration and was unable to complete your request.

The issue is related to the fact that when you declare a function to return a BLOB then you are responsible for returning the content-type headers. However, web.pb does simple datatype validation before your function is called. If a datatype mismatch occurs then web.pb returns an error message to the web server, and it needs to include a valid content-type header since your function never got the chance to return one.

Error: Internal Runtime Engine failure. Contact Powersoft Tech. Support. (NOOP).

boolean myarray[]

myarray[1] = true

myarray[2] = true

myarray[3] = true

myarray[4] = true

myarray[5] = true

if myarray = {true, true, true, true, true} then

messagebox("Hi", "hi")

end if

WORKAROUND: Check each array index value individually.

PER DEVELOPMENT:

Cannot reproduce in current builds.

Example Code:

Open (w_response) // with or without parent - same problem.

run("c:\winword\winword.exe")

ExecRemote("[fileopen ~""+"c:\somedoc.doc"+"~"]","Winword","System")

PER DEVELOPMENT:

Developer: Tested on 6.0 also and found different behavior: Word comes up faster so the execremote is in effect attempted too early. User will need to redesign app if they go to 6.0, waiting for Microsoft Word to be fully up and running.

Developer: This bug was a development time bug that showed up when adding a new external user object without specifying a dll or one that could not be loaded successfully. Sizing the susbsequent placeholder window resulted in frame garbage and tranparency.

WORKAROUND: Use LoadLibrary("PBCMP050.DLL") SDK call before opening window containing VBX.
//Global Externals Required
function uint loadlibrary(string filename) library "kernel.exe"
subroutine freelibrary(uint filehandle) library "kernel.exe"

//Instance variable for handle
uint pbcmphandle

Script to pre-load DLLs:
// just before vbx interaction
pbcmphandle = loadlibrary("pbcmp050.dll") /* loads the following files */

PBCMP, BGR, DWE, OUI, ROI, RTF, SHR.

Script to unload DLLs:
// At close of sheet/application or right after loading.
freelibrary(pbcmphandle)

Use FreeLibrary() to decrement the instance count once the window is open (so exiting the application will unload the DLL(s)) or the DLL(s) will remain in memory.

VBXs generally have lower priority.

dec ld_in, ld_out

ld_in = 10.22

ld_out = 11.22

DECLARE sp_bsel PROCEDURE FOR sp_bsel

@R = :ld_in,

@op1 = :ld_out Out using sqlca;

Workaround (less precision and not always recommended):

Declare var's as floats.

i.e.

create PROCEDURE sp_bsel_b @R float, @op1 float output

AS

begin

select @op1 = @R

end;

PER DEVELOPMENT:

Works as designed.

Works in 5.0.03 gold and Beta 6.0.

This works OK with Netscape FastTrack (using the CGI interface)

Workaround:

To fix the problem:

1. In the CGIKeywords entry in PBWEB.INI, add ",AUTH_USER" at the end

2. In all the remote methods, accept both REMOTE_USER and AUTH_USER. If

the REMOTE_USER is null then assign it the AUTH_USER. If both are null, reject the caller.

You need to edit the pbweb.ini file. There is a section at the top for CGI keywords that are allowed to be passed to the DPB server. If the variable is not listed in ini file, it will not be passed. There are 3 lines at the top for CGI, NS-API and IS-API.

Product Support Engineering

The problem in the issue is not us. O'Reilly has a few minor bugs with CGI variables. They sometimes don't map correctly. The ones that are known about are:

REMOTE_USER -> AUTH_USER

HTTP_COOKIE -> COOKIE

It's also believed that REMOTE_USER/AUTH_USER is only passed to the CGI program if the user has been authenticated. This is another issue ...

You can use a batch file to help determine which cgi environmental variables are available to Web.PB. You would envoke it from a browser.

set of=%OUTPUT_FILE%

echo Content-type: text/plain >%of%

echo. >>%of%

set >>%of%

This should be in the dos shell script directory of the web server

From O'Reilly company: WebSite 1.1 was licensed to Powersoft for inclusion with the Internet Developer's Toolkit over a year ago under an "as is" OEM licensing arrangement. It was clear to the Powersoft staff who negotiated that deal that O'Reilly is not under obligation to remedy any incompatabilities between the two products which Powersoft or its customers might identify after the release of the product by Powersoft. O'Reilly does not currently plan to issue additional service releases for WebSite 1.1

beyond those which are available from our web site for download.

11/21 - Verified fixed in v5.0.03, v5.0.04-67 and v6.0.0-beta 3

The text we're speaking of is the "Pie Chart Labels" which normally defaults to some percentage numbe (39.5%) displayed at the end of an arrow for each slice of the pie.

The example here is a 2D pie chart. Postscript and PCL drivers show the same behavior.

[Sybase][ODBC Driver]Syntax error: near '' in ..., @a_groupid = '11'[])

Workaround: Has changed the owner name from dbmaint to dba.

Turned out to be a double, in the DWE code for DDDW's, that was used before it was initialized -- making its value garbage, and under various conditions, causing an invalid double value to cause a floating point exception, which caused a GPF. Other GPFs are msvcrt40.dll.

32 bit and 16 bit {pb 5.0,5.0.02_11,5.0.03_35,5.0.04_39}

Keywords:: DDE, Voice recognition, hotlink,

Problem description::

Cannot start a hotlink when a channel is already open and another hotlink was assigned previously.

This DDE application opens a channel and uses a hot link to a Voice Recognition server. The server picks up various commands that you train it to do. Speaking the word maximise for example into the microphone will maximise the screen.

The application in Powerbuilder will work with no problems as long as there is a hotlink set up with a default user. There is an option in the app to change to a different user and perhaps use the same command or different commands. But no commands will work when you change the user.

The only way to get it to work as a different user is to close the app and re-open the channel and then do a hotlink.

It appears only one user can be use the app when a channel is open. The customer has provided a C program aswell (.exe) which demonstrates that when you change the user the application will still understand and respond to various commands, unlike the behaviour in Powerbuilder.

As per issue 395806

"The customer is attempting two conversation

> simultaineously against the same window, which uses the same window

> HANDLE,app, and topic. This is not allowed, because it is not unique.

> A DDE connection is identified as unique by these four variables:

> * appname, topicname, client window handle, server window handle.

>

> Varying any of the four variables will give the connection uniqueness,

> therefore making it acceptable.

> Think of it this way: If you have two Openchannels, which is two

> wm_dde_initiates, how does the server, in this case Excel, keep track of

> which connection it is talking to, or when a CloseChannel is issued ,

> which connection is closed. There is no way to identify which

> connection is to be closed, and which is to be kept open.

> Any attempt to do an openchannel with a connection that is already

> established, fouls up ALL existing connections between the two windows,

> causing termination of those existing connections.

> NOTE: This includes the termination of any existing Starthotlink that is

> active between these windows.

> StartHotlink is similar in that it can only have one window(handle)/data

> location to another window(handle)/data location at a time.

>

> It probably works with VB because the are different client window

> handles, or different server window handles involved in the

> conversation. Each sheet in excel is a separate window."

12-9-97

(1)If you take out the notify done from the hotlink alarmin the w_child window 'notify done' will not appear but if you click on the title bar or somewhere on the window then no text will appear in output box. Usualy ' ?? ' would appear. He claims that because no text appears then the DDE link is also broken

12-9-97: yes I verify that if you take out the 'notify done' in the hotlink alarm then run the app there is a communication between the client and server, but as soon as you click on the title bar no trained words will appear. The communication link is broken.

PER DEVELOPMENT:

Customer has changed some code in the voice recognition server engine to get it to work with his pb app. He has plans to migrate to 6.0 and will be using Web.pb and other new features of 6.0 with this application. Until then, this issue is being put 'on hold', he will cb if this particular problem exists using 6.0.

Developer: Scatter graphs only work with numerics. If you use datetime, it only recognizes the date portion. This will be documented in 6.0.01 GA Help and in the Users Guide.

DSE: user could not connect

SMI S040 039060020 Client call returns error

DWS 039060020 Exit server thread 0:0

SMI S070 039060001 Transport.StopListening()

DWS Quiesce! exit server thread TID: 0 STOP_LISTENING:14

DSE StopListening: application(oops), location(localhost), driver(WinSock) SUCCEEDED

SMI S050 039060001 Invoke Transport.Listen()

DWS 039060001 Server RPC

DWS Socket Error at line: 205, TID: 0, errorcode: WINSOCK_SOCKET_ERROR,

WinSock LastError: <<

DWS 039060023 Server RPC return

DWS 039060023 Exit server thread WINSOCK_SOCKET_ERROR:CC

DWS 039060001 Create server thread - 039060023

DWS 039060001 after Server RPC wait

SMI S060 039060001 Error: x'CC' starting Transport

DSE Listen: application(oops), location(localhost), driver(WinSock) FAILED

Looping on OpenTab()/CloseTab() using Syntax 2 for Opentab causes a GPF in PBRTF050.DLL.

Using the ImportFile(), Sort() and GroupCalc() functions. App migrated from 4.0.0x to 5.0, customer noticed significant change in execution time using these functions. In 4.0.0x 16 & 32bit, the ImportFile() function importing 15,000 records would execute in approximately 30 seconds. In pb 5.0 16bit it now takes approximately 50 minutes. The Sort() function would execute in about 30 seconds in 4.0.0x. In 5.0 16bit it now takes approximately 5 minutes. PB 5.0 32bit execution times are not as long as 16bit, but longer than 4.0.0x 32bit.

From the Table Properties window, select the Primary Key tab. There is no box to enter the name of the primary key. This results in a default name that is assigned but this creates a problem if you try to modify or delete it. It doesn't seem to know how to find it.

- primary key creation and deletion

- foreign key creation

- drop column

Results in -201 syntax error has occurred

NOTE: PFC uses this method.

NOTE: You must compile with 5.0.03 or higher. It's not enough to just 5.0.03 DLLs with an executable compiled with 5.0.02 - it will still GPF.

PER DEVELOPMENT:

PB does not guarantee the order in which events are fired.

You can create tables with columns whose datatype is timestamp and whose columnname is not "timestamp" through other means (pipeline for example) and do not have to follow this rule - seems like we should treat it as only a suggestion in table painter as well.

If you have a table which has a blob column opened in the table painter and you RMB click on the table, the pop-up menu has an option to take this table's definition into the pipeline. If you access the Pipeline Painter through this method, your blob column will be included in the "pipe-able" result set. This is also the way it works if you access the Pipeline Painter through the pipeline icon on the PainterBar

If you access the Pipeline Painter through the ToolBar, you will be asked to name the table you want to pipe. When you name a table that has a blob column in it, that blob column will not be included in the "pipe-able" result set. So you can select all the columns but not get the blob column - this requires a separate step inside the Pipeline Painter to add the blob column to the pipeline definition.

A similar problem has been discovered for combo-boxes (dropdownlistboxes) in datawindows. If a code table for a numeric field contains a number with a decimal point, that number will be misinterpreted after changing decimal separator symbol.

Per Development: This is not a bug.
When you have specified an initial value with a decimal character under one locale and try to use it with another locale and another decimal character specification, PB has no way of knowing that we must translate it to the new locale. Example; you save 1.50 in the DW under English (US) we know it is 1"decimal_point"50. If you use the DW in another locale which uses a comma for decimal point and period for thousands we now see it as 1"thousands_seperator"50. You would see a similar problem if you saved 1,200 in English(US) and you would get 1"decimal_point"200.
We simply cannot know if the number was saved under another locale.

NT: PB050.EXE - Exception: divide by zero(oxc0000094), Address: 0x11170cbd.

Workaround: Tab properties, check Ragged Right checkbox.

Cannot recreate using 5.0.04 or 6.0

Cannot recreate using 5.0.04 or 6.0.

Developer Not A Bug: Correct Scripting, etc:

/* Set a RANGE of cells...

Cannot use the ["B2:e5"] syntax because strings can't be indices.

Cannot use the ("B2:E5") syntax because Excel97 is different from Excel95

Using the "Range" method.

*/

xlApp.Application.ActiveWorkbook.Worksheets[1].Range("B2:E5").value = "My String"

Cannot recreate during 6.0 BETA.

Developer: There appears to be some behavioral differences in the windows subsystems between NT and WIN95. Both OpenChannel and/or StartHotlink initiate in the same fashion (WM_DDE_INITIATE). Multiple instances of either/both kinds of calls will break under NT 4.0. This is a Microsoft subsystem problem. The workaround is to manage the connections or use different windows for simultaineous connections, between the client and server.

Please note that this discovery applies to any 32 bit environment, NT or Windows 95.

In Microsoft's 16 bit subsystem, it was possible to establish multiple DDE conversations simultaneously between two windows.(client and server) This could be demonstrated with a simple DDE application conversing with an Excel Spreadsheet. Excel has long been considered a benchmark for DDE validation.

However, in Microsoft's 32 bit subsystems this functionality has unfortunately changed. More than one conversation CANNOT be established simultaneously between two windows, and doing so produces unpredictable results.

Because Powerbuilder calls these DDE API's provided by Microsoft, Powerbuilder is in turn constrained by these changes in the 32 bit subsystem, and therefore CANNOT support more than one simultaneous conversation between two windows. Affected Powerbuilder function is Openchannel (and there are potential problems with StartHotLink)

Customers with 16 bit applications, that are being ported to the 32 bit environments are advised to augment their applications to manage their windows, or keep conversations between windows sequential. Please note that if either the client or the server application are 32 bit code, similar problems will result, both the client and the server must remain 16 bit if retention of this functionality is desired.

The customer is tyring to modify the properties of this user object on the window and it has no effect. You can change the properties in the user object painter.

However if I create another ocx ( ctmeter ) and place it on the window there is no problem. So it is the user object inherited from CT_meter that causes the problem. Other user objects that contain custom controls appear to have the same bad effect.

Sometimes, it takes 3 or more abnormally client terminations to recreate the behavior and cause the server to hang. When closing pb on the server application, it GPF'd

The instruction at "0x10b16d45" referenced memory at "0x0000004". The memory could not be read.

WORKAROUND: The transport object variable in the example is global. If you make it an instance variable on the dpb server window, the problem doesn't occur.

Cannot reproduce 5.0.03 build 22 and beta 6.0 build 114.

In the dw painter, if you change the 'Unit of Measure' from inches to centimeters, save the dw and then go back into dw and properties, the 'Unit of Measure' will go back to inches. Tried same scenario with dw styles such as Tabular, FreeForm and Group and this does not happen. If you change the 'Unit of Measure' to centimeters and go back into the dw, it is saved.

If I scroll the window either to the right or down, and then scroll back using either the thumb or the arrows, I end up with pinstripe-like lines on the background of the window. (When I'm scrolling vertically, the lines are horizontal;when scrolling horizontally, the lines are vertical.)

If I scroll the window by clicking in the scrollbar, but not on the thumb or arrow, I don't get the lines.

I've had times where sometimes the lines are a darker color, and other times where the lines are a lighter color. They are colors that are in the pattern of the RGB.

PER DEVELOPMENT:

Windows NT handles the alignment of the dithered brush and the scrolling. I tried to override this with SetBrushOrgEx with no luck. This is not serious enough to pursue further.

Error 179 (rc -1) : Can't use the OUTPUT option when passing a constant to a stored procedure. Similar to bug 31591

Development: This was actual a fixed by a fix to the porting layer from Bristol technologies. After extracting the patch from the .tar file on Unix, do the following:

cd $PBDIR/windu/lib.sol2

cp libgdi31.so libgdi31.so.orig

tar -xvf pt400247.tar <<< this should write the new version of libgdi31.so

The sum value for libgdi31.so should be: 12672 951

You can determine if you have the correct file by doing: "sum libgdi31.so" and it should return this number. If it doesn't, then something went wrong.

Expression " gf_boolean_param(TRUE) = 1 " produces the error message.

WORKAROUND:

Modify the datatype of the function argument as noted. Alternatively use an expression that evaluates to a boolean. Using the above function declaration, " gf_boolean_param(1=1) = 1" worked fine, likewise substituting 1=0 in place of FALSE.

PER DEVELOPMENT:

Dw expressions do not support (nor do we document them supporting) TRUE and FALSE keywords

"-761 Informix Server does not match either DBservername or Dbserveralias"

NOTE: In order to connect to multiple databases, you need to use the setnet32.exe utility in either Win95 or NT 4.0

Example of object instance variable pasted by the Paste SQL and the INTO variable dialog:

select emp_id

INTO :uo_mytest.is_string

from employee;

C0014 Undefined variable.

WORKAROUND:

- Create a global function called showhelp()

- For the helpcommand enumerated type argument, define it as the type ANY.

- For the typeid use the any data type also

- in the function do a case statement

ShowHelp(HelpFile, HelpCommand, TypeId) - sample code

Arguments: HelpFile, string, value; HelpCommand, any, value; TypeId, any, value

string ls_KeyWord

long ll_TopicId

CHOOSE CASE HelpCommand

CASE Topic!

ll_TopicId = TypeId

run("hyperhelp " + HelpFile + " -c " + string(ll_TopicId) + " -s 0 -d 28395 -display : 0.0" )

CASE Keyword!

...

CASE Index!

...

END CHOOSE

RETURN 0

Developer: Definitely broken in 5.0.03 but works fine in 5.0.04.47.

Retrieve dw, rowcount > 1, then retrieve again, rowcount = 1, SetRow to row 1 fails with a -1 return code. This only happens when result set is 1 row, and previously result set was more than 1 row. Regression. Problem did not occur using 5.0.02

PER DEVELOPMENT:

Duplicate of issue 45339

on Win95

SQL0007N 'The character " " following "<table>.<column>", is not valid. SQLSTATE=42601

We're basically clipping the newline delimiters somehow during the conversion. No problem with the ODBC driver.

- drop column

Results in -201 syntax error has occurred

Using OpenLink Driver

Developer: Modifed DeleteColElement grammar in the INFORMIX_SYNTAX section

Fixed in pbodb050.ini

Developer: This is caused by the fact that PB 32 and 16 bit are registering window classes with the same name, the 16-bit code successfully registers the classes and the 32-bit version fails. Because of this, when the 32-bit version tries to create a window, it starts executing 16-bit code and that is where the problem is. The solution is to change the window class name to includes a "32" in the 32-bit version. The danger is that any test tools (Segue or SQA) or applications that are dependent on the window class name (i.e. WinSDK FindWindow()) will break.

NOTE: Powersoft has spoken to the test tools vendors and they agreed to a non-destructive fix that we have implemented in 5.0.03 and 6.0.

Specific steps to recreate:

dd/mm/yy:

1) enter 310197

2) tab to next row

3) shift-tab back

4) try to enter 311297. It will skip the zero digit, and you cannot change the date.

When I tried 280297 and 310397, 300497, and 310797 these dates all worked fine when I tried to change the month to twelve, it is not simply a problem changing single digit months to double digit months.

mm/dd/yy:

1) enter 013197

2) tab forward to next row

3) shift-tab back

4) try to enter 123197. It will again skip the zero digit and you cannot enter the 12. However, if you use the left arrow key to move back to the beginning of the date, and try to enter 123197 again, now it will let you. This did not work for the dd/mm/yy editmask.

WORKAROUND:

You can either "swipe" the whole date with the mouse to highlight it, and then you can enter the full date, or you can "swipe" just the month digits to highlight those, and then you can enter the correct two digit month. No code workaround known.

A huge memory leak if expression inside Find() function uses string function with format mask. Example: "string("+ ls_column +", [general]')"

"Could not load the plug-in 'Powerbuilder Netscape Plugin Dynamic Link Library' for the Mime type 'application/datawindow'.

The DataWIndow Plugin Requires another file called "OLEPRO32.DLL" to be present in the users Windows System directory

WORKAROUND:the workaround is for the server to not shut down the threads between remote calls from the client. To do this you use an undocumented options setting on the transport object on the server. The option string should contain "dedicated=1". This causes the test pbl to work as long as you DO NOT USE TRANSACTION POOLING. If you use transaction pooling then you will run into this bug eventually no matter what.

PER DEVELOPMENT:

Problem is related to SQL Server 6.0 using either 6.0 or 6.5 client drivers in a multithreaded app that makes cursor calls. "Freeze" is actually the MS DBLib calls looping inside of an enter critical section call. Problem goes away with 6.5 Server. May want to see if it goes away using any of the SQL Server 6.0 service packs. I noticed they did fix a number of threading issues in the service packs.

Developer: This is not a bug. This is as designed, although it's not documented fully. The application needs to guard against nested calls.

Workaround: One work around is to use API calls (setcapture & releasecapture).

work in datawindow painter when datawindow is previewed: no columns show up.

Could not reproduce in 5.0.03_25. Build sync issue.

PER DEVELOPMENT:

DOCUMENTATION CHANGE:

Add a note to documentation on SetItemStatus, Powerscript Reference Guide.

Something like this:

-----------------------------------------------------------------------------------------------------------------------

If you use SetItemStatus to change an existing row to New! and the columns to NotModified! and then change one of the columns, all of the columns in that row will change status to DataModified!

The way the DataWindow update logic works, is that if the row status is New! or NewModified!, PowerBuilder will generate an INSERT statement which includes the values of all the insertable columns in the DataWindow regardless of the column statuses.

For example, in the situation described below, if we have columns:

id, name, address

and we retrieve:

101, "Dave", "561 Virginia Rd."

now change the row status to New!, then change the the name field to read: "Dave Smith" and upon doing DW.Update(), the SQL generated is:

INSERT INTO emp VALUES(101, "Dave Smith", "561 Virginia Rd.")

It would not make sense to generate:

INSERT INTO emp VALUES(NULL, "Dave Smith", NULL)

since there was data in the columns already.

The reason why the case where the row is inserted works as it does, is also logical since initially the row values are all NULL. So, after changing the id column via SetItemXXX or via keyboard, you truly will have a row:

101, NULL, NULL

As expected, row status is NewModified!, and column statuses are DataModified!, NotModified!, NotModified. As a result, the SQL we gen:

INSERT INTO emp VALUES(101, NULL, NULL)

which indeed is indicative of what the column values are.

Fixed in PS Ref doc sources. Will be fixed in help and online books for PB 6.0.01 GA. Revised usage description of SetItemStatus function.

RAD 02/28/97: NOTE: I could not test this on 5.1_103 because I get a stack fault trying to connect with SYT to System 10.

MS Office 97. Error trying to access server function ' cell ' from Excel 8. Some methods in Excel 8 are now being treated as properties.

Customer is getting a "Powerbuilder Application execution error ( R0032)

Application Terminated. Name not found calling external object function CELLS ..."

This confirms the letter sent from Microsoft: According to attached information from Microsoft. Many Methods including calling Cell values, are now Properties instead of Methods.

This was also backpatched into 4.0.07 (Build 23).

NOTE: 4.0.07 final was build 21, so customer's will not have this patch unless they ask for it.

(once I got a more complete error message with the name of the pbl)

Not exactly a bug, but would be nice if the product handled this better since it is natural that during the development cycle this sort of change might be made. Workaround, or really correct implementation method is to add the control to the ancestor object instead of the descendant object.

Object2.DYNAMIC FUNCTION Foo()

Should give some type of Cxxx error - cannot access private function.

****This works fine in 5.0.03, 5.0.04 and 6.0

Regression Issue: Issue was fixed in 5.0.03 and was supposed to fixed in 6.0, but tested in Build 114 and was still broken. Checked to make sure updatable was checked etc. Same pbl being used.

The package is not installed.

Fontfamily drop down list.

1) emp_id_1(0) manager_id_1(0) emp_id_2(10) manager_id_2(20)

2) emp_id_1(20) manager_id_1(30) emp_id_2(10) manager_id_2 (40)

Either GPFs when trying to associate the datawindow with the control (has to insert 2 blank rows to render the display) or when opening a window containing this dataobject already placed in a control.

Customer gets an error 'attempt to free resources' in the console window when running with IDT. The customer is using PBCGI050.EXE with O'Reilly WebSite.

Everything he does gives him a gpf/access violation message pbcgi050.exe caused faults C0000005 in moduld pbcgi050.exe @ 014f:1000600c.

Also gets an ' smi40 ' error message when using pb 5.0.03_25

Possible solution to this problem is

If you take a message out of the sytem error event then PB will not terminate... This is the main problem here. Also try coding " transport.options = "Dedicated = 1". This will cut down the amount of threads used to 1.

Solved with distributed powerbuilder dlls from pb 5.0.04_65. They include pbsmi050.dll, pbdpb050.dll,pbdse050.dll,pbnpc050.dll,pbnps050.dll,pbosc050.dll,pboss050.dll,pbwsc050.dll,pbwss050.dll

Example: table t1 created by user scott. Public synonym t1 created on scott.t1. User joe creates datawindow with this syntax in DBA notepad or dynamic datawindow: select col1 from t1. Datawindow is not updateable.

On rpc calls which use setnull() to pass in null string input parameters:

sqlcode=-1

sqldbcode=999

sqlerrtext= A SQLServer request resulted in bad return code or status but no error message was returned.

If the string is passed in to the rpc call as "", it works without error.

PER DEVELOPMENT:

verified that this is NOT a bug...

(1) Customer needs to make sure that the path specified in the session.set (open it up in word or notepad etc) is valid.

(2) need to make sure that /s is used in the run command. /s stands for silent install

(3)In the session.set file c:\silent\PBDKNTI.INS should exist and that permissions are granted to get to the file.

"li_integer = lnv_attributes.TRIGGER DYNAMIC FUNCTION nf_get_integer( aany_object )" and gets back the error message "bad argument list in the function."

Determening how this is a bad argument list or not is imppossible to determine as the static function call of the same routine causes no error like this.

Running this through the debugger does not show me anything about how the argument list could be wrong. If the argument list was called statically, there is no problem. The results and returns codes etc in the debugger are exactly the same from both static and dynamic.

A nested report does not display data for the 2 record and beyond.

PER DEVELOPMENT:

Fixed by fix for bug number 42295

You can join tables together from the PainterBar but you cannot use the Menu 'Design.. Joins' this is grayed out or disabled.

Development: This problem is architectural. The DDE functions rely on the setting of internal window handle to indicate which window to do communications from. This window handle is only set when a window message is received. Because these scripts are triggered/posted, they do not go through normal message processing, the internal window handle is not set, and therefore communication is attempted with the wrong window handle. One possible solution is to create versions of the DDE functions at the object level so that the window handle is taken from the object instead.

PER DEVELOPMENT:

no longer an issue, works correcly in 5.0.03.29

Customer opens a window that contains the ct_date ocx. Gets a pbshr050.dll gpf.

(fix requires PBCMP050.DLL as well)

Size array: : returned size

----------------- ------------------------------------------------------------------

1- 8021 : 1 - 8021

8022 : 8021

8023 - 8034 : SMI call error

8035 - 12028 : 8030 - 12023 (5 bytes less)

12029 : 12025

12030 - 12041 : SMI Call Error

> 12042 : PowerBuilder just disappears

WORKAROUND: Add function to determine what array size will be and break the call into multiple calls returning portions of the data. Call the function in a control loop. Example, if you need to return 12000, call function with flag to return the first 6000 values, then call again to return the second 6000 values, putting them together in the client application.

This is very similar to bug number 41163 and may turn out to be a duplicate bug. 41163 deals with arrays of structures so it is reported separately.

Stored procedure created on MS SQL Server 6.5 database using DBMS=MSS. OUTPUT parameter in procedure declarations. If the script containing the declare is modified in any way, after saving the script, you will get an access violation/page fault either right after the save or when you try to run the application.

PER DEVELOPMENT:

appears to be a bug in mapi32. Wrote a windows app (ms windows; c code) that just loaded the mapi dll logged on and off; and freed the library . No PB involved at all. Win32 C app had same degree leakage.

WORKAROUND:

the problem occurs when the datawindow gets focus, AND the first column to get focus is an editmask required column. Try the following:

Declare a boolean instance variable on the window, boolean ib_workaround

// initialize instance in the open event of the window. You know the first time into the

// datawindow you want to avoid the error message

Window open event:

ib_workaround = true

// When you insert a row and click on the datawindow the itemfocuschanged event fires.

// This must be triggering the itemerror event. So suppress the error the first time in:

Datawindow ItemError event:

if (on the editmask column) and (ib_workaround) then

ib_workaround = false

return 1 //suppresses error message

end if

// Problem occurs when datawindow gets focus, so reset the workaround variable

// because the next time dw gets focus you want to suppress the error again.

Datawindow Losefocus event:

parent.postevent("bug_workaround")

// back on the window, declare a user event called "bug_workaround" mapped to one

// of the pb custom event ids.

Window "bug_workaround" event:

ib_workaround = true

Since the test case is very simple, you may need to do some additional checking. For instance if there are other non-required columns that could be clicked on instead of the required editmask column, you will want to check which column you're on in the GetFocus event of the datawindow and set the ib_workaround variable to false if you're not on the required editmask column, etc. Refer to Application Techniques manual for additional information on data validation.

Has a large tab control inherited from PFC u_tab with several tabpages on the tab control inherited from u_tabpage. Put the control on a sheet, save the sheet. Re-open the sheet, if you click on a tab page you gpf: PB050 caused an invalid page fault in

module PBRTF050.DLL at 0137:10d3dba6.

"attempt to convert data stopped by syntax error in source field".

All queries that do not include a timestamp work fine.

Keywords:: Web PB. incorrect URL syntax. GPF or web server crashes

Problem description::

Main issue here is that when the customer enters in an incorrect URL that errors occur. When she changed a parameter in the URL. and passed argument into a function, the whole server crashed. Essentially a gpf.

When using iis, pbcgi050.exe and internet explorer, received the following error message:

The instruction at "0x10001a56" referenced memory at "0x00145000". The memory could not be written

The web server can hang or when using iis, netscape browser and pbcgi050.exe, received the following error message:

PBCGI050.EXE Application Error

The instruction at "0x10001a56" referenced memory at "0x00145000".

Customer uses the following syntax

"dw_1.object.data.primary = dw_1.object.data.original " and this causes rows in the dw to become obscure/non readable.

However if he uses the script in the issue description you see the workaround.

per development:

cannot recreate using 5.0.03 build 35

Workaround: In MouseMove event of control add script: SetPointer(HourGlass!).

WORKAROUND from customer:

I've found that the following notation works:

for li_cnt = 1 to upperbound(gb_array)

gb_array[li_cnt] = ib_array[li_cnt]

next

PER DEVELOPMENT:

Could not recreate with recent 5.0.04

We tried to view the row through the dbadmin painter through the data manipulation grid, and we got the same results. The text that is > 2048bytes, comes back as an empty column, where as the text that is <= 2048bytes is retrieved without a problem.

32 bit only {pb 5.0, pb 5.0.02, pb 5.0.03, pb5.0.04}

Keywords:: OLE automation, Visual C++ (version 5), pointers to methods, VB

Problem description::

Basic problem is that powerbuilder does not appear to handle pointers to methods when using an ole object. Customer has created a dll using MS VC++ ver 5. He has both PB5 and VB5 applications. The VB app connects fine and calls to the ole object functions work. The PB app connects but calls to the functions fail.

Error message he gets is

"PowerBuilder Application Error (R0033) Application Terminated. Error: Invalid parameter type calling external object function assignvariantlong at line 7 in clicked event of object

cb_assignvariantlong of w_genapp_sheet."

Fixed in 5.0.04 (build 16) and 6.0 Beta (and 4.0.07 after it went gold).

ODBC driver - 6MB

O73 driver - 3.9MB

IN7 driver - 3.4MB

SYC driver - 500KB

SYT driver - negligible

MSS driver - 40KB

requiring a winsock.dll be available under Win16. The fix is created a new

DLL, pbwsd60w.dll which is for Win16 systems only. Please note the listed

DLL's in Ptrack pbwsc60.dll and pbwss60.dll DO NOT exist and should be removed.

Click on the edit property of a time column

Then change the edit Mask: to hh:mm am/pm or hh:mm:ss am/pm

Type in 12:00:00 and tab away, it is changed to 00:00:00 am

That is the bug.

However if you use the full datetime data type and type in the date aswell as the time and enter 12:00:00 as the time, tab away and there is no problem.

In NT 4.0, if you use the CONTROL-ALT-DELETE to bring up the "Windows NT Security" dialog and click on the Cancel button any window which contains a dropdownlistbox begins painting hundreds of times until it stack faults. During this time, the machine is completely locked up.

Another way to do it is to put focus in the editable dropdownlistbox edit fields and right mouse on a blank area of the desktop and go into the Properties dialog onto the "Background" or first tab page and change the "Pattern" listbox from, say, (none) to 50% grey and click on OK to apply the change. It seems this forcing a repaint on the NT 4.0 background (or Win95) causes the looping and it's NOT VIDEO CARD specific or resolution specific.

If you turn "Allow Editing" off on the dropdownlistbox, the weird painting stopped happening and it doesn't loop and GPF.

Workarounds:

1. Turn allow editing off on the dropdownlistbox

2. Use a datawindow with 1 column instead of a dropdownlistbox.

Similar to bug 44295.

Development: Fixed for NT 4.0. Win95 still freezes.

Customer is experiencing a memory leak on the client on the size of the blob. When they do a SelectBlob to a local variable, then it will chew up the memory which is equivalent to the size of the blob"

On the System tab page for a Sybase Net Gateway database profile, there is a field named System Owner. It appears to default to SYSIBM, but if you look at the profile in the preview tab page, the SystemOwner='SYSIBM' value is not being written. So when you connect with this profile, you will be prompted for SystemOwner despite having provided this on the Profile Painter.

UINT foo = 1

UINT bar = foo

We are getting compile error "C0085 Invalid variable declaration initialization", ignoring the compile error and then GPFing trying to evaluate the expression. This will happen during a full rebuild or doing a full rebuild while creating an executable.

The workaround (set UINT bar = 1) as well.

SYC (CTLIB). PB 16 bit. After 20 - 30 connects/disconnects the application terminates with the following gpf: "pbstb050 caused a gpf in module pbsyc050.dll @0002:08f0". SYB(CTLIB) terminates with gpf: pbsyb050.dll@0001:0936.

Customer is getting a gpf in pbmss050.dll at 0002 : 0003224 when using PB 5.0.03 and PB 5.0.4 16 bit . He was in the library painter trying to regen a window that had a stored procedure declared.

Note: The gpf does not occur if the customer uses the 5.0.02 version of pbmss050.dll.

The line of code from the stored procedure

" declare myproc procedure for @return_val=pef_byroyalty2 "

is causing the problem. In pb 5.0.02_11 there is no problem under any platform. In pb 5.0.04_21 the problem still occurs.

If you comment out the above line, there is no problem.

This GPF also occurs when you try to declare the same stored procedure in Powerscript. The real problem is when using the Powerscript "Paste SQL" Icon, you get an ILLEGAL script for "declare procedure" missing the procedure name - but it compiles correctly.

In 5.0.03, this doesn't compile correctly (GPFs) in regen or compile where it should give a message "syntax error: Procedure name missing from statement". If the syntax is corrected, it compiles and runs in 5.0.02 and 5.0.03.

Workaround: Fix the syntax so it's legal but this is tough since the regenerate bombs PB before you can get a list of "declares" errors. Should perhaps use the object browser.

WORKAROUND:

=============

Open composite dw, change Units of Measure property on dw to PowerBuilder units.

In 5.0.02, the GPF occurs at 0001:0C47

In 5.0.03, the GPF occurs at 0001:0C93

Error 5701 Database context changed to xxx

OLE Automation with MS Word 8.0, following script gives error: Powerbuild Application Execution Error (R0033) - Application Terminated - Error: Invalid parameter type calling external object function insertfile at line 10 in clicked event....

This is the line of code that fails:

word_obj.selection.insertfile("c:\my documents\doc1.doc")

select dateformat from master.dbo.syslanguages where langid = @@langid;

This will happen if you've changed your date format for us_english or if you are using a different langid as the default language for your SQL Server.

Customer is doing a RowsCopy() and getting a gpf in pbdwe050.dll @0044:0000044c.

This seems to happen if there are > 56 rows being copied, anything less works fine or if timestamp is not included or dw style is not GRID. Tabular and FreeForm dw's WITH timestamp do not gpf.

The syntax should be corrected to agree with Powerscript 5.0 Reference documentation on page 118.

This is what is currently painted:

Declare myproc PROCEDURE FOR @int_var...

Notice there is no "actual" procedure name after the keywords "PROCEDURE FOR". This worked in 5.0-5.0.02 because the compiler filled in "myproc" under the covers from the alias name to the actual procedure name.

NOT DATABASE SPECIFIC.

Seems to have been fixed by another bug.

module KERNEL32.DLL dragging from one datawindow to another datawindow.

Running app in 5.0.03, page fault occurs in PBSHR050.DLL@0137:1111198e. It happens on the second pass of retrieving into the ds. If you use dw's instead, the page fault does not occur.

Using dot notation to change the 3d property of a checkbox on a dw does not compile.

dw_1. modify ("column_1.checkbox.3d=no") compiles and works, but

dw_1.object.column_1.checkbox.3d='no' will not even compile.

It's because we made a change going from 4.0 to 5.0. It should work as shown above.

Development: Powerscript doesn't allow identifiers to start with a number. In this case, use: 'dw_1.object.column_1.checkbox.threed' dot notation instead if want to do this in dot notation. Doc issue as well.

OK in 6.0

Problem encountered with a datastore. If you use a GetText() to retrieve data from fields in a datastore and a SetText() to put the value into a datawindow column a validation error occurs because it is trying to return the date from the datastore to a string field in the datawindow.

Workaround: Use an invisible datatwindow.

32 bit and 16 bit {pb 5.0.02, pb 5.0.03, pb 5.0.04}

If you place additional uo tab pages within a uo tab control, place the tab control on a window and click on the tab control a few times you get a Quincy error box and PB hangs(or a gpf in PBRTF050.DLL @ 0137:10d3e096).

This only happens in development (and only in 32 bit PB. . Customer has this problem on NT 4.0 and I can recreate it on Win 95 with Pb 5.03. aswell

Fixed in pbctl050.dll 5.0.04.26. ALso need to include pbroi050.dll from pb 5.0.04_26

Hangs on placement of non-PowerSoft OCX

System hangs when I place an non-PowerSoft OCX on a window.

Any of the microsoft OCX controls such as combo box or forms cause pb to hang in any of the above versions. These OCX controls come with office 97.

With Win NT 4.0 you cannot resize some of microsoft controls at all (multimedia player can be resized and does not crash)

datawindow yyyy editmask on a date column. If 2000 is entered then 0000 is displayed.

Private:

int test

If you then declare a user event and later inherit u_decendent from u_ancestor and try to call the u_ancestor event via

super::event eventname()

in u_decendent, you will receive an error:

"C0109: Private or Protected function cannot be accessed:"

when you try to regenerate the u_decendent object. The PRIVATE instance variable declaration in u_ancestor seems to trigger this. If you also have PUBLIC: in the same instance variable area of u_ancestor, the error doesn't occur when regenerating u_decendent.

For instance, if a DDLB Text property is populated in the Constructor event, it is shown in the DDLB. However, if a function calls for the DDLB Text property, it is not extracted right away.

Workaround: Use unbounded array.

Developer: Problem occurs because you are trying to pass a bounded array to something that takes an UNbounded array by reference. Since it's by reference and the types don't match, it's a problem. This is fixed in 6.0 but cannot be in 5.0 due to the severity of the fix

PER DEVELOPMENT:

Cannot reproduce using 5.0.04 or later

The datawindow is generating:

WHERE ... AND created_datetime = {ts '1997-02-03 17:48:00.00000'}

Which causes the error:

SQLSTATE = S1000

[INTERSOLV][ODBC Informix driver][Informix]Extra characters at the end of a datetime or interval.

This should be the syntax it's generating. This works.

WHERE ...AND created_datetime = DATETIME (1996-12-11 09:45:00.00000) YEAR TO FRACTION(5)

Development/Product Support: 5.0.04.71 or 6.0.01 build 349.

32 bit. {pb 5.0.02, pb 5.0.03, pb 5.0.04 }

Keywords:: MS sql server 6.5, sybase sys 11. MSS, SYB, dynamic SQL

Problem description::

Execute Dynamic does not return result sets when using the MSS driver in NT 4.0. It will return correct result sets when using the SYB driver.

The customer is trying to get to the same stored procedure which sits on MS sql server 6.5 using differnent drivers (which are supposed to function the same way) but do not.

You need to use Dynamic SQL for the problem to occur. If customer statically swiches from syb to mss and vice versa there is no problem. Its only at run time that this problem occurs.

The Sybase sql server version 4.X is similar to MS sql server 4.X, but it is not similar to SQL server 6.5 when using dynamic sql.

Workaround: Check Multiline and Ragged Right properties of ancestor tab and save object.

Database connections used in the example: profile name = MSS 6.0, DBMS = MSS, database = pbdev1, server = sql95, ID = t1user, password = t1pass profile name = SYC-system10, DBMS = SYC, database = qadata, server = system10, ID = qalogin, password = qapass

To recreate through Pipeline Painter:

1. Open the Pipeline Painter. Click new.

2. Select SQL Select as Data Source. Use Database Connections: Source is MSS 6.0, and Destination is SYC-system10.

3. Select any table. Select all columns. Click SQL.

4. Execute. Works fine.

To recreate through Database Painter:

1. Through Database Painter, connect to MSS 6.0. Open the same table as #3 above.

2. Click Pipeline icon. I used the option "Create-Add table".

3. Click Destination Profile icon. Select SYC-system10. Click Connect.

4. Notice that t1user is appended to table name.

5. Execute. Error message 2734 "User name t1user does not exist in Sysusers".

6. Redo steps 1 through 5, but this time use SYC as the Source and MSS as the Destination. Receive error message 2710 "You are not the owner specified for object named dbo.table_you_chose in this command (CREATE, ALTER, TRUNCATE, UPDATE STATISTICS, or BULK INSERT). The database owner can use the SETUSER command to assume the identity of another user."

Similar errors occur with Oracle and Informix database connnections. With the Informix connection, a "." is appended to the table not the ID.

PER DEVELOPMENT: If something is passed to BlobEdit as a blob, no terminating NULL is included, whereas it is if the item passed is of the string type. At a fundamental level a string is just a bytestream with a terminating NULL, so this makes sense.

A datetime column in a datastore that has an editmask of mm/yy/dd on it will not print. If he takes the editmask off and uses the edit style it prints fine. Works fine in 5.0.02. If he uses a datawindow instead of a datastore works fine in 5.0.03.

32 bit { pb 5.0, pb 5.0.02, pb 5.0.03, pb 5.0.04, panther 6.0 beta 3 version 189}

Problem description::

If the customer tries to insert a wmf file that he created from win nt 4.0 with a dw.SaveAs function into an ole container ( Word 7.0 or 8.0 ) he can not see it.

If you save the metafile from win 95 then just activate it from nt 4.0, you will see no problem. This proves that the dw.saveas() is causing the problem.

We tested this out with word 7.0 and word 8.0.

Also tested with NT 4.0 service pack 3 and nt 4.0 with no service packs.

The results were exactly the same.

On some machines running win nt 4.0 800X600 RESOLUTION causes the problem while

1024X768 RESOLUTION works fine

A work around is to subtract 2 from the pointer argument before the BlobMid function is called, then add 2 back to it, if you're keeping track of the pointer.

PER DEVELOPMENT:

Not a bug. gf_raise is used to calculate the datawindow computed object salary. Every time you go from the debugger to the datawindow, the focus changes which causes all computed objects to be updated which calls gf_raise ....

Development: Drawing objects are always in background.

32 bit only { pb 5.0, 5.0.02_11, 5.0.03_35 5.0.04_41}.

Keywords:: Dynamic tab pages, treeview, keydown!

Problem description::

The main problem here is that when the customer scrolls through a list in the treeviewitems and holds down the CTRL key and the down arrow, that the top level treeviewitem is deleted, then subsequent treeviewitems after it.

The item will reappear if you run the app again.

Note Ctrl-Left and Right Arrow scrolls one pixel when a TVI is selected. Why does this happen. Also shifting by one pixel to the left and right happens with CTRL and left/right arrow key. This should not happen

Other Questions::

Question 1: Why does Ctrl + Arrow keys allow scrolling when there is no need to scroll? (Note Ctrl-Left and Right Arrow scrolls one pixel when a TVI is selected)

Question 2: How can default tree keypress behavior be intercepted and/or overridden?

Question 3: How can I control horizontal tree scrolling? I already know about VScroll using Send(Handle(This), 277, x, 0)

Develpment will look at Question 1

Question 3 above is related to issue 421025 which is still open in the dev-gui Queue.

Question 2 look at issue 455981, 423387 some info in here about it. Place a mesasgebox or a return of some kind. Or else trigger the error event to stop this from happening.

Application error R0035 at the InsertFile() command trying to insert into an Word OLE Control on a window. Using the Clear() command.

Example Code:

The customer is using

olecontrol.InsertFile("temp.doc")

olecontrol.SaveAs("c:\example.doc")

olecontrol.Clear()

olecontrol.InsertClass("Word.Document")

olecontrol.Activate(offsite!)

olecontrol.object.application.wordbasic.InsertFile("c:\example.doc")

This worked fine in 5.0.02.

Error received: R0035 message- Error Calling External Object Function 'Cells'. Underlying this error is one coming from Excel 5.0 from using PB's internal traces:

...GetProperty Invoke returned 0x00000000

...GetDispID - GetIDsOfNames returned 0x80010002 << here

PBROI_Term()

PBROIApplication:: dtor

NOT A POWERBUILDER BUG:

This issue also happens in Visual Basic 4.0 16 bit and 32 bit. The script that duplicates the error:

The same thing (OLE2) on Visual Basic 4.0 gets the same error after 47 iterations.

Script under a command button going against the Excel 5.0c Worksheet on a Visual Basic form.

Dim count As Integer

For count = 1 To 300

Label2.Caption = count

OLE1.object.cells(1, 1).Value = 1997

Next count

The runtime error in Visual Basic 4.0 16 bit under WFW 3.11 using Excel 5.0 16 bit.

Runtime Error: Ole automation error: -2147418110 (80010002)<<<< Notice same 80010002 error from Visual Basic 4.0.

The runtime error in Visual Basic 4.0 32 bit under Win95 using Excel 7.0 32 bit

Runtime error: Ole Automation Error: -2147023174 (800706ba). Excel 7.0 dies at 545 iterations.

NT 4.0 seemed to be a little bit more robust and VB 4.0 32 bit and Excel 7.0 32 bit completed 3000 inserts.

The Microsoft issue # is SRX971009600309.

Other reserved words also have this problem such as "update[]", etc.

The script compiler catches this.

User reports that ProfileString is not correctly matching if a section header includes an equal sign surrounded by spaces.

This is probably not a bug, we probably just send the call right to GetPrivateProfileString and that is not handling the spaces the way the user expects, but I didn't find documentation on GetPrivateProfileString that specifically addressed this scenario and user wants this reported as a bug.

Here is what the profile.txt file looks like:

<beggining of file>

[section 1]

key=Banana

[section = 1]

key=Apple

[section=1]

key=Orange

[section =1]

key=Grape

[section= 1]

key=Guava

<end of file>

In Powerbuilder, here are the function calls and what is returned:

Profilestring("c:\profile.txt","Section 1","key","Failed!") >>> Returns Banana

Profilestring("c:\profile.txt","Section=1","key","Failed!") >>> Returns Apple

Profilestring("c:\profile.txt","Section = 1","key","Failed!") >>> Returns Failed!

Profilestring("c:\profile.txt","Section =1","key","Failed!") >>> Returns Failed!

Profilestring("c:\profile.txt","Section= 1","key","Failed!") >>> Returns Failed!

--------------------------------------

PER DEVELOPMENT:

The GetPrivateProfileString function in Windows does the space mangling you are seeing. If you really need to care about spaces inside key names, you will need to embed quotations. Personally I recommend against it as Microsoft is no longer actively supporting INI files. However the following information from the Knowledge Base may be useful:

-----------------------------------

An application can retrieve information from a profile file by calling the GetProfileString or GetPrivateProfileString function. If the profile file associates the specified lpKeyName value with a string that is delimited by quotation marks, Windows discards the quotation marks when it copies the associated string into the application- provided buffer.

For example, if the following entry appears in the profile file:

[application name] [application name]

keyname = 'string' or keyname = "string"

The GetPrivateProfileString and GetProfileString functions read the string value and discard the quotation marks.

This behavior allows spaces to be put into a string. For example, the profile entry

keyname = string

returns the string without a leading space, whereas

keyname = ' string' or keyname = " string"

returns the string with a leading space.

Doubling quotation marks includes quotation marks in the string. For example:

keyname = ''string'' or keyname = ""string""

returns the string with its quotation marks -- 'string' or "string".

When closing word I get a GPF "PB050 caused an invalid page fault in module rpcrt4.dll at 0137:7fd07317.

The next GPF I get is " GPF in KRNL386.EXE "

Customer is sharing data between two datastores. RowCount() from secondary datastore comes back with 0 (zero), it should be 2.

Development: This particular issue was fixed when other OLE window redraw issues were fixed.

Customer has an MDI application. He opens a sheet, and pops a menu with a right mouse click that opens a response window.

He then right mouse clicks the reponse window which pops up another menu.

Once he closes the reponse window, the menu on the frame does not work. You can navigate the menu, but no scripts are executed when a selection is performed.

This is due to a problem with a subselect statement in the pbsyc2.sql that comes with either pb 5.0 or 6.0. Pbsyc2.sql changes the syntax for the sp_pb50table stored procedure created in pbsyc.sql so the user sees only tables that he has access to. You can use the original sp_pb50table (or sp_pb60table) stored procedure that comes in the pbsyc.sql file by rerunning the pbsyc.sql file or if you want to fix the pbsyc2.sql file the following change can be used:-

Original statement:

or ((select max(((sign(uid) * abs(uid-16383)) * 2) + (~(protecttype/2)))

from sysprotects p

where p.id =* o.id

and (p.uid = 0

or p.uid = user_id()

or p.uid =* u.gid)

Change to:

or ((select max(((sign(p.uid) * abs(p.uid-16383)) * 2) + (~(protecttype/2)))

from sysprotects p,sysobjects q, sysusers r

where p.id =* q.id

and (p.uid = 0

or p.uid = user_id()

or p.uid =* r.gid)

Dot notation gives also an uncomplete syntax. The "edit.case = lower" is missing in the resulting syntax.

Solution Per Development:

This issue has been defined as an enhancement and not as a bug. The interface was developed in accordance to the Netscape Enterprise 2.0 NS-API interface. Therefore Powersoft's statement currently is that we don't support the NS-API 3.0 interface. This same situtation happens each time a database vendor provides an new version of their database drivers. Development has committed to and is working on a NS-API 3.0 interface.

Customer is handling pointers to methods when using an ole object. Customer has created a dll using MS VC++ ver 5. 2 gpfs occur when closing an application involving the invokation of variant and long arrays

On win 95 Error "PB050 caused an exception 03H in module KERNEL32.DLL at 0137:bff76694."

On win NT 4.0 " " The exception Breakpoint has been reached. (0x80000003) occured in the app at 0x77f75d64."

10-15-97

Development says that the return coming back from the server is of unknown memory size. So the user will need to know ahead of time what memory to allocate or pass unbounded arrays.

In win NT 4.0 get a Dr. Watson

An application error has occurred and an application error log is being generated PB050.exe.

Exception: access violation (0xc0000005, Address 0x10b14472)

In win 95

the error message there is:

PB050 caused an invalid page fault in

module PBRTE050.DLL at 0137:10b14472.

Solution Per Development:

This can't be fixed in 5.0 because it requires an architectural change. The library list has to be correct. In 6.0, no GPF occurs (but there is no helpful error message). This type of message is already given during incremental rebuilds, etc.

In 6.0, the GPF is in module PBVM60W.DLL at 002c:00000101.

NOTE: The original bug to fix window class collisions bug 41416 - is still OK. 41416 was strictly window class name collisions. This particular bug occurs when there is a datawindow on the particular window(s) in the 16 and 32 bit application.

Per Development:

This issue cannot be addressed in any maintenance release or EBF. This is because changing class names will affect existing applications that are dependent upon the original class names.

Development: This is a documentation issue. Registry functions all refer to RegistrySet for detailed usage info. RegistrySet documents that HKEY_CLASSES_ROOT is the only root string accessible on Win 3.1.

1950-02-15 becomes 2000-01-01 in the "date" portion when you're only trying to change the "time" portion.

The bug perhaps here is that since the date value is in the buffer, why generate a bogus date on something already retrieved.

If you leave the date portion off by editing the "sqlpreview", you get todays date

UPDATE "timestamp" SET "timestamp1" = '12:15:35.000000'

WHERE "newid" = 1 AND "timestamp1" = '1950-03-15 12:10:00.000000'

becomes

January 8, 1998 12:15:35 - the date this issue was entered.

If you let it default, the system always generates 2000-01-01 for the date portion.

Customer is using an OLE/OCX gamesman control. The control is CTyear. He was originally using pb 5.0.02 and the control worked fine. When he went to 5.0.03, they were having trouble with ole/ocx controls.

You get an error about external function daycolor not being passed the correct number of arguments.

Drag and drop: dragging from a dw and move to another dw which happens to not be a valid drop dw, it changes to the invalid drop icon, but when moving cursor away from the invalid drop dw to a valid drop dw, the icon never changes back and it starts to flicker then a page fault occurs: PBRTF050.DLL@0137:10d1f67f (in 32bit). In 16bit, it boots you out of powerbuilder and windows.

- Error "CLI0125E Function sequence error. SQLSTATE=S1010" (for DB2/MVS)

- SQLState=S1000 Connection is busy with results for another hstmt (for MS SQL Server)

Development: Needs the dot notation or Modify() after the fact to change the moveable attribute. A dynamically created datawindow might not have these and definitely not if it isn't a grid.

Note: no SharedObjectGet() is involved

Workaround: Use Triggerevent()

Probably closely related to child window with titlebar bug 41443. As with bug 41443, HOME/END seems to unhighlight the edit control so users can edit.

DW plugin does not show up in Internet Explorer; in Netscape after hitting the reload button two page faults occur: NPDWE60.DLL@0137:00ebb2fc, click OK another page fault in MFC40.DLL@0137:5f80103b.

Comments: 16 bit on NT 3.51 only !

Step-By-Step Instructions (In development environment):

1) go into the Application painter

2) right mouse click , selecting properties,

3)-> the tab folders are "white-ish" without any 3d effect.

PER DEVELOPMENT:

The tab control functions correctly in both 5.0 and 6.0. The above test scenario can't be used to verify the tab control behavior, because MessageBox calls disturb the message control flow.

To verify that all constructors are being called please replace the MessageBox calls with e.g. writing to a file. You will find that the constructor events for selected tab pages and controls on the selected tab page are always triggered even though CreateOnDemand is TRUE.

When creating a pipeline based on this table, the column appears as a NUMBER(43,5). This produces an error : "numeric precision specifier is out of range (1 to 38)"

If you step through the source code in the second session and place a break point

and step via the debugger in assembly language you will get a message that an Access Violation error occured. The title bar will read " PBCMP060.dll " which indicates the dll is broken. All other dlls that are read in to this point have no errors attached.

WORKAROUND: Using the pipeline to re-create the table seemed to clear up the problem.

Development: - If you unload of database, and examine the data contained in the .dat files, you will see that the "date" column "fccreacion" is of type date, but the data is actually of type datetime. Now, this would not cause a problem if all the time portions were set to 00:00.

(I believe that all dates are stored as datetimes internal to the db). In this case, however, there is a specific time stamp associated with some of the rows. If you go through the .dat file and see which rows have the 00:00 time stamp, you will find that the "row changed between retrieve and update" message will not occur. For example, try to change/update row 1 ( '2BDAT', -1 ). There will not be a problem. I believe that when you pipelined the data, it must have stripped the time ( zeroized ).

An example of an inserted value that might "take" and get stored in the database as a datetime:

'95-12-01 00:00:02'

The PFC-Specific GPF:

gpf in pfcdwsrv.DLL 0137:02a1276e

This is not the same as bug 40759 which was fixed in 5.0.03 regarding a GPF right at a getitemdecimal() call going to a NULL decimal column.. This bug specifically deals with returning a NULL decimal cast to an ANY datatype variable and GPFing right at the "RETURN <any variable>" statement. This is exactly how of_getitemany() FORMAT 4 is structured in PFC "pfc_n_cst_dwsrv". PFC uses a non-visual user object function but this can be recreated this with a window function.

NOTE: ALL other numeric datatypes work in machine code with NULLs using of_getitemany() - datawindow types ULONG, LONG, REAL, NUMBER all work correctly.

This bug exists in 5.0 and 6.0

Workaround: (for PFC users)

First option

=========

Make the calls to Getitemdecimal() manually rather than using of_getitemany(). You could evaluate the "Describe("#1.coltype") and only if it returns "decimal", use Getitemdecimal() manually with in-line script.

Second Option

==============

Object: pfc_n_cst_dwsrv

Function of_getitemany() format 4 - returns ANY

CASE....

...

CASE "decim"

la_value = dw_1.GetItemDecimal ( al_row, as_column, adw_buffer, ab_orig_value )

// Add some checking code here

if isnull(la_value) then

decimal ld

ld = "-100"

la_value = ld

// cast this with a NON-null but identifyable value so it doesn't GPF at the 'return' call

return la_value

/*----------- end of of_getitemany() function ----------*/

/* back in powerscript - trap for your decimal NULL condition*/

decimal ld_decimal

setnull(ld_decimal)

if of_getitemany(x, y) = -100 then

setitem(a,b,ld_decimal)

end if

DEVELOPMENT: Fixed in compiler in pb 5.0.05 and later. Requires regenerating your machine code exe.

Solution:

Please refer to the documentation for String() in the Return value section. It specifies that it returns and empty string if the data type of the data passed is an incompatible type. Enums are not compatible w/ this function

Solution:

This is not a bug. As indicated in the 6.0 documentation you should not call any visual function from a shared object or server session. On Unix they will not work. See above for more detail.

See PSE notes in issue for details.

DOCUMENTATION ERROR:

The documentation and help should refect this.

1) The clicked event on the dw will fire first if there is code in it. The return code here is unchanged from 5.0 and has no effect on the new button events. Return 0 allows normal dw processing to continue (ie, column clicked on gets focus), Return 1 stops the focus from changing.

Note it is not a good idea to use messageboxes in this event. It can inhibit the button events that follow.

2) There are two new events in the dw; buttonclicking and buttonclicked. These will be fired as described below. Also there is a Suppress Event Processing property for the button object. If it is on, the buttonclicking and buttonclicked events are not fired, but the action assign to the button is executed.

3) If the Suppress Event Processing property is off, the buttonclicking event will fire after the clicked event (if the clicked event has code), and before the action assigned to the button is executed. A return code of 0 here allows the action to be executed followed by the buttonclicked event. A return code of 1 will inhibit the action from being performed, and inhibit the buttonclicked event. So processing here prior to the action can be used to inhibit the action from executing.

The order of events then is:

Suppress Event Processing ON:

---------------------------------------------

1) clicked if coded

2) button action executed

Suppress Event Processing OFF:

---------------------------------------------

1) clicked if coded

2) buttonclicking

3) button action executed (if buttonclicking returned 0)

4) buttonclicked (if buttonclicking returned 0)

[n] <--- This means transparent when set from the property sheet. This is not available from the dropdowntoolbar.

Migration of FTPTest.PBL with SOCKETS.PBL omitted causes gpf in pbvm60.dll

Migrated an application from the AppGlry\FTP\FTPTest.pbl. I accidentally did not include both pbl's (omitted sockets.pbl from library list) and received a gpf in pbvm60.dll. (in build 220 I received a gpf in pbcmp60.dll)

Has a function that retrieves the sql statement appropriate for the particular db connection and it calls a dynamic cursor that fetches one record. They call that function that executes the dynamic cursor approximately 5000 times, it gpf's between 1,000 - 2,000 iterations in module PBSHR050.DLL @ 000D:0799

Window Open event

Postevent("ue_timer")

Window ue_timer event

timer(1)

This works fine if you use a triggerevent instead of a postevent.

GPF in PBVM60.DLL at 0137:0053d745 when trying to insert an OLE control and there are no registered controls on the machine.

Appears fixed by bug 44892 fix.

Each time that you enter any of the development painters, handles are allocated but never released. Run the etsk manager and notice the amount of handles being taken up but not released.

DEVELOPMENT: could not reproduce.

Fixes uploaded to: //ftp.powersoft.com/pub/private/pbpatch/48091_32.zip (for 32bit) or 48091_16.zip (for 16bit)

Developer: Cannot put out a warning message -esp with datastores. Perhaps some kind of computed field which would show "DATA TRUNCATED" or null for a visible thing.

(PFC_N_CST_DWCACHE, of_refresh (string as_id)).

any la_args[20]

..

// nulls in all entries

ds.retrieve(la_args[1], la_args[2]......) <-- GPFs here.

Line 123 of of_refresh(string as_id) contains this construct in the

CHOOSE CASE

statement "RETRIEVE,SQL" section. The PFC examples uses this construct populating the listview and treeview on w_examplemain - the frame window for the examples.

Workarounds: For pfc pfc_n_cst_dwcache service of_refresh (string as_id) function.

Case RETRIEVE, SQL

// Workaround: Check to see if null arguments exist in the first few arguments

// of the la_args[] array of any that is used in the retrieve(). If they are null,

// this means a 'normal' retrieve for a datawindow with NO retrieval

// arguments - exactly the example that causes the GPF IF YOU USE AN ARRAY.

// instead, catch the specific null condition.

if isnull(la_args[1]) and isnull(la_args[2]) and isnull(la_args[3]) then

// 3 entries is enough. Don't feed it the array if all nulls. This causes the GPF.

// instead do your own retrieve()

li_rows = istr_cachelist[li_cnt].ds_obj.Retrieve()

return li_rows

else

// normal processing with old PFC code. At least one any array argument is non-null.

// This doesn't GPF.

li_rows = istr_cachelist[li_cnt].ds_obj.Retrieve(la_args[1]....la_args[20])

return li_rows

end if

NOTE: This should be fixed soon, but if you need to use the workaround, as always, don't edit the PFC_ layer, add the function on the PFE_ layer where it will override the ancestor automatically.

ds.Object.DataWindow.HTMLTable.generatecss="yes" causes the error R1041 "Error accessing external object property generatecss

The .HTMLTable.NoWrap property will also fail.

Workaround: Use Modify() instead.

ds.Modify("datawindow.htmltable.generatecss='yes'")

dw_1.Object.datawindow.HTMLTable.NoWrap = 'yes' // This will fail

dw_1.Modify("datawindow.htmltable.NoWrap='yes'") // This will work

PER DEVELOPMENT:

the following syntax will work:

dw_1.object.datawindow.htmltable.generatecss = "~"yes~""

Per Development:

Fixed only in PB7.0 because it was a major change that could not be fixed in a PB 6.0 maintenance release.

Mouse events do have x,y. OCX_y , OCX_x but drag events don't reveal x,y

Development: The VCI doc refers to the VisualBasic "DragDrop" event which is different. Documentation deficiency.

------------

2/3/98 JMartin

The problem is the registry entries that are created by the PB 6.0 installer. The entries are documented in the PTrack.

The Installer program needs to be modified so that these entries are not created.

------------

The error from IE 3.0 when loading a page with a pbd embedded in it states "An error occurred inside a plug-in contained on this page".

The result from accessing either the window or DW plug in from IE 4.0 is that the area/rectangle is blank (no icon) and if you right click on the document you can analyze the doc. The result of the analyze is: "The following EMBED did not install properly." It goes on to say:

This might have been caused by one of the following conditions:

1. Your current security settings prevent this EMBED from being used.

2. This EMBED was not propertly installed on your computer

3. The page or EMBED was authored incorrectly.

Environments that fail:

Window Plugin: IE 3.0, IE 4.0

DW Plugin: IE 4.0

NOTE: See bug #48588 for the status of the Datawindow Plugin fix.

PER DEVELOPMENT:

Solution:

PB Development spoke with Microsoft Tech Support - they recommended to:

**completely uninstall IE4.0 then re-install IE4.0

**There is a technote at Microsoft website about how to _completely_ remove IE4 from a client machine.

Also - make sure that the browser security options allow _any_ plugin to run.

Also - like the discovery above, make sure the PB5 plugin is not being loaded rather than the PB6 plugin.

Also - IE will look into the NETSCAPE 'plugins' directory when looking for the plugin to load...

Also, when 2nd connect fails during SelectBlob processing, PB GPFs.

2/9/98 Update:

============

File located: ftp://ftp.powersoft.com/pub/private/pbpatch/48726_32.zip.

Contains PBSYC050.DLL from 5.0.05.

Three main enhancements added:

1) New static DBParm lets the user override the default CS_MAX_CONNECT property for the Client Library context used by PBSYC.

Usage: sqlca.dbparm="maxconnect=nn" where nn = number of connections you wish

to allow for this context.

2) New dynamic DBParm, OptSelectBlob=1, lets the user optimize SelectBlob processing

by using the current transaction object to perform the SelectBlob request. It is the user's responsibility to make sure that they are not doing any datawindow retrieval of blob or binary data while OptSelectBlob=1. This dbparm can greatly improve performance for applications that use the SelectBlob powerscript function. User can specify OptSelectBlob at connect time or at any other time while the connection is open. The

following example shows how to turn OptSelectBlob for one specific request:

sqlca.dbparm="OptSelectBlob=1"

SelectBlob blobcol into :myblob_variable from blob_table where id = :idvalue;

if sqlca.sqlcode = 0 then

...

end if

sqlca.dbparm="OptSelectBlob= 0"

3) Fixed GPF in PBSYC that occurred whenever the 2nd connection requested by Select Blob processing failed.

1/9/98 RENEEN PSE Update:

Regarding the backspace behavior, per development:

"In the edit mask code we DO intend to keep the '0' if we are backspacing over it's position in the mask ... maybe not "correct" behavior but it looks like original code."

This was verified with PB 4.0.03 gold and 4.0.07 gold, PB 5.0 gold through 5.0.04_70, and 6.0 gold (build 251) and 6.0_332. As the developer points out, this may not be "correct" behavior but it's original code in the product and changing it at this point could cause problems for other customers that have built applications based on the current behavior.

>>I do not feel that we should change this behavior at this point because of the impact it may have on other customers.

Regarding the editchanged event not firing:

>>This event now appears to fire using PB 5.0.04 Build 70 and 6.0 gold as well as 6.0_332.

Regarding the editchanged event returning the wrong value:

>>I feel that this is the only behavior that should be investigated for a fix.

Per Development:

Partially fixed. The error where we might return incomplete data when the mask has a comma in it is fixed. The behavior of the delete key and what is returned to editgettext in the itemchanged event when the delete key is pressed is not changed. Too much risk of screwing up other applications.

Placing Visual Components Formula 1 ocx v.4.1.0.5 on a window in NT 4 will cause intermittent hanging, Dr Watson and sometimes pb will simply crash without warning.

------------

2/3/98 JMartin

The problem is the registry entries that are created by the PB 6.0 installer. The entries are documented in the PTrack.

The Installer program needs to be modified so that these entries are not created.

------------

NOTE: See bug #48389 for the status of the Window Plugin fix.

PER DEVELOPMENT:

PB Development spoke with Microsoft Tech Support - they recommended to:

**completely uninstall IE4.0 then re-install IE4.0

**There is a technote at Microsoft website about how to _completely_ remove IE4 from a client machine.

Also - make sure that the browser security options allow _any_ plugin to run.

Also - like the discovery above, make sure the PB5 plugin is not being loaded rather than the PB6 plugin.

Also - IE will look into the NETSCAPE 'plugins' directory when looking for the plugin to load...

A bitmap opened on a window, as the window is resized, there is alot of flickering. Also if you open another window ontop of another and then close the top one, the other window loses it scrollbars. If you click in the area where they should be the scrollbars repaint.

Error 242 "implicit conversion from datatype varchar to int is not allowed use the convert function to run this query"

Workaround: Prefix stored procedure with owner name, ie dbo.spname.

February 9, 1998 - StephenD (Stephen Dupre) PSE

The REAL problem: 6.0.00 Regression in 16 bit. Works fine in 5.0.04 16 bit and 32 bit and 6.0 32 bit. A 16 bit only problem with casting a numeric into an any variable which incorrectly evaluates to >1 and <= 0 - an impossible condition. Most numeric datatypes have the problem.

// here is the example.

int li_int = 1

long ll_long =1

decimal ld_decimal = 1

double ld_double = 1

real lr_real = 1

any la_rc

la_rc = <ll_int, ll_long, ld_decimal, ld_double>

if la_rc < 0 then messagebox("evaluate < 0", "true")

if la_rc >= 1 then messagebox("evaluate >=0", "true")

Any of the variables shown will return "true" in both sets of messageboxes. It should only return true for the "evaluate >=0" messagebox.

For a REAL datatype cast into an ANY variable under 16 bit, it seems NEITHER if ... then statement evaluates to "true"

Again, all numeric datatypes cast into any variables under 32 bit work (even REAL)

The very specific 16 bit PFC problem

==============================

PFC has a construct in the pfc_n_cst_luw (logical unit of work service)in the object function

of_updatespending (powerobject apo_control[]) returns integer

that evaluates a "1" assigned to an any variable incorrectly as a -1 in an if..then statement as a window with pending changes is being closed.

//excerpt from of_updatespending((powerobject apo_control[]) returns integer)

any la_rc

la_rc = lpo_tocheck.Dynamic of_UpdatesPending ()

If ClassName(la_rc) = 'integer' or ClassName(la_rc)='long' Then

// Functionality was found.

If la_rc < 0 Then Return -1 /*** << both evaluate to true

If la_rc >= 1 Then /*** <<

//some code

end if

The dynamic call above to of_UpdatesPending() during a window close with pending changes returns "1" into la_rc in both 16 bit and 32 bit (as seen in the debugger) but when doing the IF...THEN evaluations following evaluates BOTH

If la_rc < 0

and

if la_rc >= 1

to true - an impossible condition. Since the "if la_rc < 0 return -1" evaluates true first, the system errors out with a bogus validation error when it should produce a "Save changes y/n dialog".

IMPORTANT NOTE:

Product Support Engineering: There are other similar constructs that are fixed with this fix. Dot notation returns an any and in an IF..THEN construct would evaluate TRUE always using 6.0.00 16 bit:

if dw_1.object.datawindow.table.select = 'total_garbage' then

Messagebox("evaluate", "true")

end if

This could have far-ranging effects so far as code execution is concerned. You must be using 6.0.01 build 366 or higher to get the patch that fixes this.

Using ##,###,###.0 If user enters negative number (ex, -60), edit mask only accepts the 6. The only way it works is to type in number and then backspace to put in - sign.

Using ###.00000

a) If user puts in .365, edit mask reads it as 365.

b) If user puts in 0.365, edit mask accepts it as .365.

Product Support: These both exist in 5.0.03 but not in later builds due to many changes in editmask processing.

Retrieval of data is successful under these scenarios:

- There is any DBParm (w/out StaticBind, which it will default to 1)

- Set StaticBind = 1 explicitly even though it is considered to be the default

- Set StaticBind = 0 and DisableBind = 1

- Set DisableBind = 1

Workaround: If using default StaticBind= 1, hard code the DBPARM where the database connection information is or in your database profile. Or set a different DBPARM (if applicable).

1) Enter the script painter for any event of any object.

2) Select "Paste SQL" from the PainterBar.

3) Double-click "non-cursor / Update".

4) Click OK to select any table.

Notice, on the "Update Column Values" window, that the child window on which you select column names is too small to fit the scroll bar completely. I remember an earlier version of PB where this child window was actually movable. This seems OK, but the right half of the vertical scrollbar is cut off. I'm using a Thinkpad with SVGA/800x600/small fonts.

This has also been reported with menus. The menu painter is unable to save script of a descendant menu if the ancestor has no script. The script moves 1 item up. If the menu shows:

File

New

Open

and you script the clicked event, it ends up in the "New" clicked script.

PER DEVELOPMENT: We do not support remote arrays, only remote objects. The object passed to the server when of_popinfo is called is a pointer to an object on the client, so it is remote. Then setting the array elements local to server.

WORKAROUND: Call a remote function to get the data in the server and assign it locally to the array.

Can't recreate in 6.0.01 build 360 in Win95 or NT 4.0.

WORKAROUND FROM DEVELOPMENT:

similar to the fix made for parsing negative numbers.

You can workaround the bug in the script by adding a " " (space) after all the commas in the script.

// ORIGINAL SCRIPT:

// Set Series Lines

string ls_avg = "50.4"

string ls_lcl = "-10"

string ls_ucl = "120"

string ls_goal = "80"

string ls_actual

// Change values property of graph

this.dw_1.Object.gr_1.Values = "yaxis,goal," + ls_avg + "," + ls_lcl + "," + ls_ucl

//CHANGE THE LINE ABOVE LIKE THIS:

// FROM:

"yaxis,goal"+lcval+","+ldval+"," .....

// TO:

"yaxis , goal" +l cval+" , "+ldval+" , " .... // add a space before and after each comma.

Workaround:

IF isValid(theCurrentFriend) AND (theCurrentFriend <> aDWFriend) THEN

// other code

END IF

The work around I used was quite simply to embed the second condition test as follows:

IF isValid(theCurrentFriend) THEN

IF (theCurrentFriend <> aDWFriend) THEN

// other code

END IF

END IF

WORKAROUND:

Set the return type to (None) or to integer (or whatever is needed) and click OK instead of Cancel. Now the correct return type is reflected when you go in to look at the event definition under the Args button.

Workaround:

Call the PostEvent from the Descendant Open Event since the event won't get fired until after it's script is finished - even when called from the Ancestor Open Event.

1) The INI file was saved as a unicode text file

2) The first two bytes of the file are 0xFFFE

3) The section name being requested is on line one, immediately following the invisible 0xFFFE.

PER DEVELOPMENT:This is a bug in ProfileString. This is true for all applications using the windows api under unicode. We can't fix this.

Possible workarounds:

1) include a routine that makes sure the first line of the INI file is a blank (either check everytime, or else create a file containing only a blank line if the file doesn't exist) - this will prevent the collision of the first section header with the Unicode flag bytes at the start of the file.

2)make sure that the INI file is stored as ANSI (the API routines automatically handle translation when required, so having an ANSI INI file presents no problem to a Unicode app) by creating the file as an ANSI file before first writing to it from the app.

Development: This will be a documented limitation starting in 6.0.01.

Compiling a user event script with a referenced argument and getting a compiler error "c0014:undefined variable <arg_name>".

Development: In old code needs regen or migrate or bug will appear.

Problem:Page fault in call to PB OLE Automation server when calling functions.

This worked correctly in 5.0.03.

Developer: Not a BUG.

When CreateOnDemand is set to TRUE the constructor event for a tab page is not called until users view the tab page. The fact that it has been called in the past (5.0) was actually a bug (which was used as a feature here). Please use the following approach to achieve the desired behavior:

//ue_postopen event

*************************

int li_ctr

string is_title

this.setredraw(false)

for li_ctr = 1 to 5

is_title = "Tab#" + string(li_ctr)

tab_test.opentabwithparm(iuo_tabpage[li_ctr],is_title,0)

iuo_tabpage[li_ctr].text = is_title // Set text here and not in the constructor!

next

this.setredraw(true)

return 1

This will be updated in the 6.0.01 documentation.

PER DEVELOPMENT: DOCUMENTATION:

Added info that columns may be truncated if report too narrow to UG p.673. Fix will appear next time online UG is refreshed on web server or hardcopy reprinted.

Goto Topic SetItem() Powerscript Function, select Syntax 1 for Datawindows, Datastores and child DataWindows. Then click Examples command button. The examples displayed are for the Sort() Powerscript Function.

This is also the case for Syntax 3 for ListView control column.

PER DEVELOPMENT: DOCUMENTATION:

Fixed in PSRef doc sources. Will be fixed in Help and online books for PB 6.0.01 GA.

When loading the psr file into the browser, it will GPF. Problem exists in the NPDWE60W.DLL. Reproduced the problem using Netscape Navigator 3 on Win 3.11

GPF occurs in PBROI050.dll when you double click on the OCX icon in the Object Browser painter on the OLE tab. The problem only exists if Microsoft Visual Interdev is installed on the system.

Development: We don't see the behavior in 6.0.00 or 6.0.01. Assume they didn't have the thing registered. (if they were dual booting and forgot to re-install under Win95, this could easily happen.

Simplist case is using %,dba (no quotes) in the System Tab of the Profile dialog which is the equivalent of:

DBPARM="TableCriteria='%,dba' "

It works in 5.0.04 but not in 6.0.00 (returns no tables)

Follow-on from or variation of 424807 bug 43003. Does not work in 5.0.04.72 or 6.0.01 either. This syntax compiled in 5.0.02.

subroutine MARLENE(long INARG1,string INARG2,ref long OUTARG3,ref string OUTARG4) RPCFUNC ALIAS FOR "~"dba~".~"something~""

The key is the "~"dba~".~"something~""

There are many ways to put quotes in aliases so there are several flavors of this bug.

NOTE: Does not work in 5.0.04. Will be fixed in 6.0.01 only.

The Yield function doesn't allow any other events to be handled. The customer has a process button that they need to be able to suspend/cancel from another button. They use the yield() method in a tight loop and test an instantance boolean to see if the cancel button was pressed. The window doesn't respond to anything and the browser is locked up.

Tested against the NS 3.0, 4.0 and IE 3.0 browsers with the same results.

Solution:

This is actually not supported because PB automation server does not support visual objects.

The Return Value should read:

"Long. Returns the number of the row displayed at the top of the datawindow control when the scroll is complete or if it tries to scroll past the last row. ScrollNextPage returns 1 with Composite reports and child datawindows since the current row cannot be changed in these cases. ScrollNextPage returns -1 if an error occurs. If dwcontrol is NULL, ScrollNextPage returns NULL."

PER DEVELOPMENT: DOCUMENTATION:

Fixed in PSRef doc sources. Will be fixed in Help and online books for PB 6.0.01 GA. Additional information added to Return Value description.

The Window Plugin (nppba050.dll and nppba60.dll) do not work on IE 4.0 when there are multiple LIBRARY properties defined in the EMBED tag of the HTML page. Testing has shown that the problem only exists with IE 4.0. It seems that IE doesn't fetch the additional .PBD files that are listed in the EMBED tag.

This appears to be a definite bug in the IE 4.0 browser.

Using the WinSock trace tool, I watched the communications of the browser to the web server. The browser is not handling the EMBED tag correctly. This is the sequence of events.

- Browser requests winplug6.htm

- Browser requests winplug.pbd (declared as SRC=)

- Browser requests extra1.pbd (declared by the first LIBRARY= entry)

- Browser re-requests winplug.pbd (THIS CLEARLY SHOULDN'T HAPPEN)

- No requests are ever maded for the additional LIBRARY= entries.

Per Development:

Bug in IE4 - PowerBuilder workaround/fix put into PB6 and later.

The bug in IE4 appears to be that IE4 does not allow multiple tags with the same keyword.

Hence - I modified the Netscape Plugin to accept:

LIBRARY1="file1"

LIBRARY2="file2"

LIBRARY3="file3"

Actually, a tag "LIBRARYanything" will work too.

Hence a correct webpage which works in both NetScape and IE4 is:

-----------------------------------------------------------------------------------------------

<HTML>

<HEAD>

<TITLE>PowerBuilder Window Plugin - Issue 499901</TITLE>

</HEAD>

<BODY BGCOLOR="#FFFFFF">

<H2>PowerBuilder Window Plugin - Issue 499901</H2><hr>

<h3>Testing problem with multiple LIBRARY parameters in the EMBED tag</h3>

<P>

<EMBED SRC="winplug.pbd" type="application/vnd.powerbuilder6"

WIDTH=550 HEIGHT=400

WINDOW=w_child

LIBRARY="extra1.pbd"

LIBRARY2="extra2.pbd"

LIBRARY3="extra3.pbd"

LIBRARY4="extra4.pbd"

LIBRARY5="extra5.pbd"

LIBRARY6="extra6.pbd"

LIBRARY7="extra7.pbd"

>

<P>

<HR>

<h3>After plugin...</h3>

</BODY>

</HTML>

-----------------------------------------------------------------------------------------------

an openwithparm was done using a response window and an autoinstantiate object as parameters AND the close come from a WM_SYSCOMMAND message (Alt F4)

Per Development:

Not a bug. The issue is that every time you step, the focus changes. This causes getfocus to re-execute and re-enter the debugger. Having a comment is the same as having just a return statement, so it behaves the same.

'~001' through '~006' and '~014' through '~031' are actually stored as '~111' (compiled machine code).

Place the following single line in an application's open script and compile it.

MessageBox('', String(Asc('~001')))

The transaction.SyntaxFromSQL method will not work in a distributed environment on the HP platform. Apparently the SyntaxFromSQL code is trying to access GDI resources and that is not valid due to the fact that the XWindows libraries are not thread-safe. This is similar to bug # 49877.

Workaround: It is possible to save a window with a groupbox, edit it's properties in exported source and re-import. Moving and resizing a groupbox does not cause a GPF. There are only a few properties in a groupbox.

The GPF occurs when SetItem() function attempts to put a datetime value in the datawindow after disconnecting client using RemoteStopConnection().

Entering 12.34, highlighting and re-entering 12.34 causes the "1" to stay around when you start typing, this causes problems and corrupts the editmask.

Problem: Attempt to check out an object using SCC API with an error causes the object to appear checked out even though it is not.

Per Development:

Normally Jaguar should only be connected to by Open Client clients and not ODBC clients. However for Power Dynamo, Jaguar catches all of its ODBC calls and deals with them. Since the ODBC driver treats CS_LONGCHARs as blobs and the ODBC driver doesn't support that, a change was made so that if a string column is 254 (NOT255) or less, it will go through Jaguar as a CS_CHAR, which the ODBC driver can handle. Also if the datastore is from an ORACLE database, ORACLE longs aren't supported as they are actully long character strings (32K) in the datastore. ORACLE longs can be used if the client is an Open Client (ctlib-based), but not Power Dynamo ODBC.

Additional Details about the 48863_32.zip Jaguar patch:

DLLs:

pbvm60.dll 2/5/98 (3382272)

pbdwe60.dll 2/5/98 (1604608)

These dlls are a special build. They were created to address a specific problem that was encountered when using PowerBuilder and Jaguar (see issue #492550 below for specific details). The dlls were created using PowerBuilder 6.0 GA, and include only the following three additional fixes:

Issue 1 - #492550, Bug #48863

Description: A GPF will occur when connecting via ODBC to a database in a PowerBuilder component running in Jaguar CTS or MTS.

NOTE: Issue #492250 was the original problem reported and the reason this patch was created. The two problems described below were identified by Development as a result of researching a fix for issue #492550.

Issue 2 - #495305, Bug #49106

Description: dw.GenerateResultSet fails with the following error from Jaguar "srv_xferdata null indicator not set when datalength = 0" for a character column that contains a blank space (vs. being NULL) in a database.

Issue 3 - #492401, Bug #48829

Description: The datastore.GenerateResultSet function does not work properly when the datastore is populated through an ODBC connection to an Oracle database for columns of type long, char and fully populated varchar2.

Occurs only under 16bit under win95. 16bit under WinNt works properly.

Application Execution Error R1020

"Function doesn't return value at line 113 in function of_UpdatesPending() of object pfc_n_cst_luw"

Fixed in 6.0.01

Per Development:

p-code should never disable and check off all the pbd's - not a bug. It never did this - not a regression. This is required and works for machine code only.

PER DEVELOPMENT:

This is not a PB bug as originally thought but a failure to send the data in the right format according to the spec.

Per Development:

This is not a PowerBuilder bug. THIS IS A BUG WITH THE MicroSoft INSTALL:

The Install for Outlook 98 correctly creates an entry in the registry for the SchedulePlus library :

CLASSES_ROOTCLSID\

{800DD100-DB43-11CE-914E-00A004000162}

which contains a number of subkeys, including VersionIndependentProgID. This is fine.

However, the installer also incorrectly creates a subkey:

CLASSES_ROOTCLSID\

{800DD100-DB43-11CE-914E-00A004000162}VersionIndependentProgID

which does not belong here. The presence of this faulty key causes the GPF in PowerBuilder.

This erroneous key can be removed from the registry with no ill effects.

PB WORKAROUND/FIX:

A PowerBuilder workaround/fix has been provided so that customers do not have to edit their registry to fix the Microsoft error.

PER DEVELOPMENT:

Not a bug. While this looks unusual and it is a novel way to use rowsCopy this is the way it works if you have overlapped your source range and before row destination.

PER DEVELOPMENT:

This is an issue with the printer drivers, not a Powerbuilder issue.

You would create the executable without checking off the PBD option for the PBL that contains the datawindow objects. All the datawindow objects have been assigned to datawindow controls so they are definitely not "dynamic".

Solution Per Development:

This is a documentation issue, not a bug.

Datawindows are included in an executable only if they are referenced by

windows or user objects that are included in the exectuable. When a window

or user object appears in a pbd, the datawindows they reference (even if

they are assigned statically) must be referenced in the pbr file (PB

Resource file) associated with the pbd. IMPORTANT NOTE: Statically assigned

icons and bitmaps work the same way.

PER DEVELOPMENT:

Cannot recreate in later builds.

PER DEVELOPMENT:

Fixed in 6.0 code base 9/29/97. Fixed in 5.0 code base 9/29/97.

PER DEVELOPMENT:

The issue is caused by the default behaviour of Novell for new files, as is not something we can change (see the full description). This should be documented, however.

Issue transferred to documentation.

Cannot recreate during 6.0 beta.

PER DEVELOPMENT: NOT A BUG:

I found 3 code changes that are required in the app's PowerScript. Let's start by reviewing the given code in the test application.

cb_delete.clicked: // the DeleteRows button on the window

int i

for i = 1 to dw_1.rowcount ()

if dw_1.IsSelected (i) = TRUE then

dw_1.DeleteRow (i)

end if

next

dw_1.RowFocusChanged: // the primary data window

dw_2.ScrollToRow (this.getrow ())

First of all, the for-loop is the wrong construct for bulk deletes. The 'i' variable increments on each loop (1,2,3,...), but if Row 2 is deleted, then Row 3 becomes Row 2. 'i' increments and tests Row 3 (which was Row 4), and the old Row 3 (now Row 2) does not get tested. The following do-loop tests each row...

do while i <= dw_1.rowcount ()

if dw_1.IsSelected (i) = TRUE then

dw_1.DeleteRow (i)

else

i ++

end if

loop

This loop only increments the row number when the tested row is NOT deleted. If Row 2 is deleted, then Row 3 becomes Row 2, and Row 2 gets tested again.

Second, each DeleteRow issued against dw_1, causes a refresh of dw_2, which could lead to timing problems, since the data table of dw_1 is still being manipulated (potential GPFs). All refreshes of dw_2 should wait until all dw_1 manipulations are complete. Therefore there should be a dw_2.SetRedraw (false) before this do-loop, and a dw_2.SetRedraw (true) after the loop. This also speeds-up the delete procedure, by suppressing all the dw_2 refreshes.

And lastly, I found that if I selected rows from the top-down in dw_1, and deleted them, I hit the GPF. If I selected rows from the bottom-up (selecting the same rows, including or excluding first and last rows), I do not get the GPF. This was related to the dw_1.RowFocusChanged event. Selecting from the top-down, the last row selected, at the BOTTOM of the DW, becomes the current row. Selecting from the bottom-up, the last row selected, at the TOP of the DW, becomes the current row.

If I select from the bottom-up, and the last row selected is Row 2, and this becomes the current row, and the first row deleted, then Row 3 becomes Row 2, and the current row remains as Row 2. If I select from the top-down, and the first row selected is Row 2, and the last row selected is Row 10, and Row 10 is the current row, then when Row 2 is deleted, Row 10 becomes Row 9, and the RowFocusChanged event fires.

The RowFocusChanged event tries to sychronize the shared DWs with a ScrollToRow in dw_2. This was causing a repagination of dw_2 (because of the deleted row), but the row deletion operation was not yet complete, and table synchronization problems caused the GPF. The ScrollToRow should be delayed until dw_1 has completed its operation.

dw_1.RowFocusChanged: // the primary data window

dw_2.PostEvent ("GoToRow")

dw_2.GoToRow: // the secondary data window; new user event

this.scrolltorow (dw_1.getrow ())

Or, better yet, suspend RowFocusChanged actions until the do-loop is complete -- again, speeding up the delete process. The resultant code would look something like this...

w_test window instance variable: // new variable

boolean ProcessRowFocus = true

cb_delete.clicked: // the DeleteRows button on the window

int i

dw_2.SetRedraw (false)

w_test.ProcessRowFocus = false

i = 1

do while i <= dw_1.rowcount ()

if dw_1.IsSelected (i) = TRUE then

dw_1.DeleteRow (i)

else

i ++

end if

loop

w_test.ProcessRowFocus = true

dw_2.SetRedraw (true)

dw_2.PostEvent ("GoToRow")

dw_1.RowFocusChanged: // the primary data window

if w_test.ProcessRowFocus = true

dw_2.PostEvent ("GoToRow")

dw_2.GoToRow: // the secondary data window; new user event

this.ScrollToRow (dw_1.getrow ())

Does an insertrow() on a dw that has a ddlb. Then a setvalue() on that ddlb, then a modify to change the background colour. If the setvalue tries to insert to a row on the ddlb greater than the next availble row, it gives a return value of 1 but gives a gpf on the modify.

This fix will not allow the user to include a datetime field in the update clause. Development has now been asked to put this code back into the product, and provide customers with an option whether or not they want to include these fields in the where clause. The default will remain as is, these fields will not be included and customers will need to explicitly override the default using the new dbparm option..)

New DBPARM for 5.0.04 and 6.0: DateTimeAllowed='yes'

Default behavior will be to NOT include datetime columns in the WHERE clause.

Workaround: (if possible) use native driver.

PER DEVELOPMENT: NOT A POWERBUILDER BUG:

SQL Anywhere has an internal sqlda structure that only holds 32 characters (which meets the original IBM spec)

However, now that the engine can store larger column names, if the odbc driver receives a column name that is 29 characters, it is supposed to be "suspicious" of truncation and query the engine for the entire column name. This was only being done in the case that the sql had been executed via SQLExecDirect or SQLExecute. In PB, we had just prepared the statement and had not executed it yet. The SQL Anywhere driver has been fixed to do the check regardless of execution state.

Fixed in wod50t.dll build 1680 5.5.0.3

PER DEVELOPMENT:

Fixed during 6.0 beta.

We had basically hard coded some returns like "login failed", etc.

PER DEVELOPMENT:

This was fixed in both 5.0 and 6.0 code base. We now give "Login Failed" and SQLDBCODE=4002 for a bad password. This happens in PBSYB, PBSYT, and PBSYC.

per development:

cannot recreate using 5.0.03.01

Workaround: Put a space between the ranges and around the dash:

1001<space>-<space>1500

The key is to make sure of the order of the columns in the table. The problem occurs if the longer named column is before the shorter one, e.g.

c_prod_no

c_prod

c_pro

We are doing a Match() instead of a equality comparison under the covers. Other more subtle things might occur as well.

You will see problems with the destination part of the Group dlg (i.e. the list on the right) no matter the order in the table - just try dragging cols into the right-side list in the order I list above; you should see c_prod replace c_prod_no when you drag it...

Similar problems actually occur in the Sort dlg as well (try dragging columns to the right-side; also deselecting cols from the table window)

instead of inserting it if focus is on the 'Type' field.

NOTE: Duplicate of 168806.

A workaround exists, ie. to manually create a "normal sql source dw" and changing the source to a stored procedure in the exported datawindow.

PBNET60 is at end-of-life and is not shipping with 7.0. Since there is a reasonable workaround this problem will not be fixed.

Notice no primary key, tab is there and you can't add columns to the table. Everything is greyed out

This error occurs because the Invalid date used for the PREPARE is constructed incorrectly.

informix.sysdistrib
informix.sysfragments
informix.sysmenuitems
informix.sysmenus
informix.systrigbody
informix.systriggers.

Per Development:
Not to be fixed.

Second leak. Property sheet of button control in dw. Cycling through "action" dddw to show icons (use default icon for action checked) was leaking slightly due to the painting of the icons (for example for retrieve).

6/8/98 Fbashaw: Verified fixed in build 439

Per Development - no bug -- works as designed. The Copy feature is something the user has to program into the application. In a DW Preview we provide the user with menus and toolbars for things like Retrieve, Update, InsertRows, SaveAs, etc., -- and Edit / Copy of a grid selection (Edit /Copy of the contents of an EditBox is handled by Windows common controls). None of these features are 'automatically' associated with a DW Control dropped on a window -- they have to be programmed by the developer.

Sometimes, when scrolling down, the GROUP HEADER doesn't paint correctly but scrolling up (with the THUMB) does print correctly

****R&D response (March 9/98) -- NO bug; NO regression; NO fix. The scrollbar operation for DataWindows was changed in 6.0. The virtual scrollling enhancement allows any row to be at the top of a screen view (even in a group dw), and print preview scrolling to scroll from the top of the report to the bottom of the report, all in one slide. This means that it is possible to scroll to a point that shows the bottom half of one page and the top half of the next page. This is what's happening in this issue. Even in screen views the data is paginated in a fixed layout. The page size of a screen view is the size of the visual dwControl. The data was arranged into groups and it was determined that a group header was required in the middle of page 6 and repeated at the top of page 7. This layout does not change because of the scrollbar position. If the arrow keys or the page up/down keys are used exclusively, then the issue does not arise. The scrollbar enhancement was added to override the navigation keys.

PB Version and Build: 5.0.04.76 ----> Fail
5.0.05.10 ----> Fail
6.0.01.419 --> Fail
7.0.81 ---------> Fail

The menu option contains the following script:

window lw_test
lw_test = parentwindow
messagebox("OPEN MENU OPTION",classname(lw_test))

The PowerBuilder online help for menu object properties states the following:

ParentWindow - Specifies the window that owns the Menu object.

Using PowerBuilder 5.0.04.76 and 5.0.05.10, when the parentwindow property is called from a command button on a visual user object using the following script:

m_menu lm_pop
lm_pop = create m_menu
lm_pop.m_file.popmenu(200,100)

The name of the "command button" is returned instead of the name of the window.

In PowerBuilder 6.0 and 6.0.01.419 the name of the window is returned correctly.

When a NON VISUAL userobject is used with the above script, all versions of PowerBuilder (5.x, 6.x, 7.x) return the name of the non visual userobject instead of the name of the window.

The following are related cases:

CR 162471 ---> Ptrack 50145 ---> Bug number 49657.

Ptrack 272083

Win95 hangs and gpfs for customer in PBDWE60.DLL at 0257:105e01fd.
NT 4.0 GPFs in pb60.exe 0xc000005, address 0x78008dfa

This worked in 5.0.04 and 6.0.00.

The Datawindow Engine/Painter needs to be rewritten in order to allow text and large character columns to be displayed accurately when it does not fit a on a physical page. There is 'a longstanding DW requirement that rows fit on a page.

Function of_GetDetail()

Description ==>> Creates an array containing references to all details for the current DataWindow.
Access ==>> Public
Syntax ==>> dwcontrol.instancename.of_GetDetails ( details[ ] )
Description ==>> Creates an array containing references of all details for the current datawindow.
Usage ==>> Call this function from any DataWindow in the linkage chain to access its detail DataWindows.

Problem:
=======
Documentation needs to include a description, and/or a warning, when using the of_GetDetail() function along with the of_Unlink() function. There was an enhancement request to enhance the functionality of the of_Unlink() function to address this issue.

A small note should perhaps be added to mention something like the following:

''When using the of_Unlink() function along with this function, be sure to check all the elements in the Linkage chain are actually referencing a u_dw object as you loop through the linkage chain. Keep in mind that the of_Unlink() function does NOT remove elements from the PFC Linkage chain. It only re-references the the element to an ramdom section in memory.''

Excerpt from enhancement request:
''Currently PFC Unlinks a datawindow from the linkage chain by creating a local variable of type u_dw and assigning it to the element in the array that references the datawindow control the user wants to unlink. This is done by calling the of_unlink() PFC function. The of_unlink() function then calls the of_unlikdetail() function which performs the re-assignment.''

Note: How PFC development will deal with issue is directly revelant to the outcome of this one.
RKN: I will work this out with PFC development.

If a datawindow has a long char column (255 char) that has Auto Vertical Scroll and Autosize Height selected when a new row is inserted and a user is inputting data there are differences in behavior from NT 4.0 and WIN 95. In NT 4.0 if the space bar is pressed multiple times at the end of the visible input line to try to take the user to the next line the cursor does not move until a valid character is entered. The extra spaces however are retained in the row and in the database. In Win95 the appropriate number of spaces hit to take the user to the end of the visible input line takes the user to the next line and no additional spaces are retained to the database.

I have managed to reproduce the access violation with the following international characters: � � � � �

I have performed the following tests with all builds of PowerBuilder 6.0.01:

6.0.251 GA - Will allow me to create a 32 bit machine code executable.

The following characters are displayed in message boxes with my test application:

� - � - Correct
� - � - Correct
� - o - Incorrect
� - � - Correct
� - o - Incorrect

If I create a 32 bit p-code executable, all the characters are displayed fine.

6.0.01.360 - Will allow me to create a 32 bit machine code executable.

The following characters are displayed in message boxes with my test application:

� - o - Incorrect
� - o - Incorrect
� - o - Incorrect
� - o - Incorrect
� - o - Incorrect

If I create a 32 bit p-code executable, all the characters are displayed fine.

6.0.01.365 - Will NOT allow me to create a 32 bit machine code executable.

I get the following link error: Codegen compilation error, see file 'D:\TEMP\test2.log'.

Test2.log contains the following information:

CGEN: Compiling 'D:\TEMP\test2.c'
D:\TEMP\test2.c(13) declaration specifiers are required to declare 'INT'
D:\TEMP\test2.c(12) syntax error before 'WINAPI'; probable cause: incorrectly spelled type name
D:\TEMP\test2.c(45) syntax error before 'INT'; probable cause: incorrectly spelled type name
D:\TEMP\test2.c(493) syntax error; probable cause: missing ';'
D:\TEMP\test2.c(498) syntax error; probable cause: missing ';'
D:\TEMP\test2.c(500) syntax error before 'VOID'; probable cause: incorrectly spelled type name
D:\TEMP\test2.c(537) syntax error before 'INT'; probable cause: incorrectly spelled type name
D:\TEMP\test2.c(2134) symbol 'VTAB_CLASS_INFO' already defined
D:\TEMP\test2.c(2134) 'int VTAB_CLASS_INFO' defined in: D:\TEMP\test2.c(493) (col
D:\TEMP\test2.c(2134) syntax error; probable cause: missing ';'
D:\TEMP\test2.c(2144) symbol 'VTAB_GROUP_INFO' already defined
D:\TEMP\test2.c(2144) 'int VTAB_GROUP_INFO' defined in: D:\TEMP\test2.c(498) (col
D:\TEMP\test2.c(2144) syntax error; probable cause: missing ';'
D:\TEMP\test2.c(2146) syntax error before 'VOID'; probable cause: incorrectly spelled type name
D:\TEMP\test2.c(2170) no reference to symbol 'VTAB_CLASS_INFO'
D:\TEMP\test2.c(2170) 'int VTAB_CLASS_INFO' defined in: D:\TEMP\test2.c(493) (col
D:\TEMP\test2.c(2170) no reference to symbol 'VTAB_GROUP_INFO'
D:\TEMP\test2.c(2170) 'int VTAB_GROUP_INFO' defined in: D:\TEMP\test2.c(498) (col

If I create a 32 bit p-code executable, all the characters are displayed fine.

I get the same results as 6.0.01.365 after testing with 6.0.01.370, 6.0.01.375

With 6.0.01.381 I get the following access violation when trying to compile a machine code executable 0xc0000005 0x00333437

I also get the same access violation while testing with the following builds of PowerBuilder 6.0.01: 6.0.01.385, 6.0.01.390, 6.0.01.399, 6.0.01.404, 6.0.01.409, 6.0.01.414, 6.0.01.419. 6.0.01.424, 6.0.01.444

If I create a 32 bit p-code executable with the above builds, all the characters are displayed fine.

In PB 6.0 GA (build 251 tested) there was a different problem, that has been fixed in PB 6.0.01, that masked this problem where if you highlighted the field and hit the delete key a huge negative number appeared in the field.

Without the editmask it triggers the editchanged event when the delete key is hit and the itemerror event when tabbed off but with the editmask on neither the editchanged or the itemchanged events get fired. If valid data is input, even a "0" then both the editchanged and the itemchanged events get fired.

Workaround: Map a user event on the datawindow called ue_key to the system event pbm_dwnkey. Code the following:

long ll_null, ll_row
int li_column
string mod_string, ls_coltype

if key = keydelete! then

// did you select ALL text for delete? Bug 166243

if len(selectedtext()) > len(gettext()) then

// what type of column am I on? Make sure you're on a column

if not isnull(getcolumn()) then

// get the row and column numbers
li_column = getcolumn()
ll_row = getrow()

mod_string = "#"+string(li_column)+".coltype"
// get the column type "deci, numb, long, ulong
ls_coltype = describe(mod_string)
// Error checking.
if mod_string <> "!" then

// get value ready
setnull(ll_null)

// find out if it's numeric
choose case left(ls_coltype,4)
case 'deci','numb','real','long','ulon'
setitem(ll_row,li_column,ll_null)
end choose

end if /* mod string check*/

end if /* getcolumn not null */

end if /* Selected full text in editmask */

end if /* Pressed DELETE key */

Win95 does not have this problem. LaserJet 4 driver.

Only with ole controls and standard visual uo of type olecontrol
Only reproduced with an OLE custom control (OK with 'Create New' or 'Create from file')
Any ole custom control will show this, example is with Visual Components Formula 1
Not reproduced with other PB controls

The same code in other events (such as a user event posted from open/constructor events) gives the correct runtime error - R1071 - and exits the window

Tested with Netscape 3.0 to 4.0.04 and MSIE 3.0 to 4.0 on NT 4.0 and Win95

There is no acceptable workaround. Could post a user event from constructor/open events but this will not make the client machines secure from other web apps which have been developed without using this alternative.

Problem 2: If user1 checks an object out into ObjectCycle, user2 should NOT be allowed to alter the checked out object, but currently user2 has the ability to do this.

There is definitely a change in behavior from PB 5.0 to PB 6.0. It appears as though PB 6.0 is interfering with the windows messaging or this could also be simply unsupported behavior.

There is a mixture of long text columns and nested reports in a freeform style but the entirety does not flow beyond a page.

What is happening is the long text columns are print previewing to a larger font and this makes it appear to overlap the nested reports below it (even if the autosize is set properly everywhere).

Per Development:
Not a bug -- This is the long standing overlap bug due to conversion from print to screen resolution.
This is another example of PrintPreview facsimiles not being able to reproduce an exact image of the printed page because of the integer-math of screen resolutions. This report has a bad layout design of a mixture of AutoSizeHeight text columns, and AutoSizeHeight nests. On the printer, the data is presented in the proper position. On the screen the initial large text columns on the detail band expand to twice their print size, because of the lower resolution -- and therefore higher font height -- of the screen. The text expands to appear under the subsequent nests. This is a long-standing side-effect of PrintPreviews of large text columns. If the PrintPreview is to be used to view the report, then the design should be changed to put each of the large text columns in its own nest. The last lines of the nests will probably get clipped -- due to accumulated line height differences -- but at least the blocks of text will not be overlapped. Also, once you start placing nested reports in a detail band, the remainder of the detail band should all be nests -- that is, do not add more columns or text fields after the nests, because of nest pagination rules.

In all versions of PB 5.0 and PB 6.0 prior to build 424 his script works fine. Starting in build 424, and also failing in build 429,
this same script gives an illegal page fault in PBVM60.dll. In build 424 the address is 0137:11218af5 and with build 429 the address is 0137:11218b05 on my machine.

This has been simplified by removing the function and simply assigning the instance NVO to the local NVO that has not yet been created in the script on the button. The same illegal page fault is generated.

Duplicate of bug 166615

6/8/98 fbashaw: Verified it works correctly in build 439.

Similar to cr163009.

with 6.0 DataWindow Error
Select Error: ORA-01460: unimplemented or unreasonable conversion requested
with 6.0.01 the error message doesn't show, but the data is not retrieved

There are also differences seen with the sample when the column has focus and whether or not it has autosize on or not. From speaking with GerryR. this is most likely just behavior and not a bug.

Works properly under PB 5.0x

August 3, 1998 - Product Support Engineering
Workaround found: Use another level of composite (put the composite in a composite) and the group report will start after the top header report without going on to the next page.

Tab Order Protect Rowfocuschanged via Setrow()
--------------- ----------- -----------------------------------------------
1. <all zero> <all protected> Fires <<- huh? Why fire it in this case but not case 2?
2. <10,20,etc> <all protected> Does not fire (setrow(), getrow() fail) <<- This is the complaint
3. <all zero> <none protected> Fires
4. <10,20,etc> <none protected> Fires

Once the datawindow is put into Print Preview, you can see the problem in the pagecount.

Also, if you take either the DETAIL band button or the HEADER band button out, this doesn't occur. It has to be in BOTH bands.

Workaround: Try turning autosizeheight of the detail band off and size the columns so they don't need slide or autosize.

ld_date = 2001-02-29 <-- This compiler bug is CR162708.

setitem(row,col,ld_date)

Development: SetItem() does not do any validation. The bug is in the compiler that allows the specification of an invalid date constant bug 162708.

Although this data has not been validated yet, by deleting the '1' in '12', we effectively let the user extract a valid date.

Development: Not really a bug. The editmask control cannot prevent all possible conditions and keystrokes. We would have to disable the delete key for this circumstance and that would not be appropriate

6/10/98 fbashaw: Fixed in 6.0.01 build 442.

This bug is a duplicate of 166615. Fixed in 6.0.01 build 442.

If you are using three or more PBDs/DLLs with the OLE generate registration feature, and if the total length of the LibraryList using a long filename path is greater than 276 characters then PowerBuilder will produce an access violation (a different address is generated each time) on NT 4 and PowerBuilder will hang indefinitely on Win 95. The problem occurs on the second press of the "Proceed" command button in step 5 of 6 of the OLE automation registry generation dialog box, usually the "Proceed" command button will work first time, but not under these circumstances.

Development: Release note stating customers should rebuild all applications to avoid Application Painter load delays when using 6.5 Powerbuilder with older 6.0.00 applications and even when running the application.

We change the internal type definitions slightly in 6.5 and this caused ANYTHING big being opened to take longer with 6.5 - meaning an Incremental or Full Rebuild (best) was required to sync up an old application with the new type definitions.

We "recompiled" pfc for 6.5. By taking an older 6.0 application opening it with this version of PFC, then PFC (6.5) is newer than their application and this causes a mini incremental rebuild on startup. If the developer uses the original 6.0 version of PFC then everything is in sync, of course.

Any time the ancestor classes are newer than their descendants (as in this case using a 6.0 PFC application with newer 6.5 PFC pbls - as just about everything will be) then the vm must do a psuedo-rebuild of the descendants since the ancestor has changed. This happens when you open the app painter because it encounters one of these situations and then does the rebuild. So think of it like you are doing an incremental rebuild every time you open the app painter (or any painter for that matter). The difference is, this rebuild does not get saved back to disk so until you do a real one incremental rebuild, it will keep happening.

Of course, the work around is to do an incremental rebuild and it is then solved.

The problem is there is nothing we can do to solve this. This is actually correct behavior and really is a staging issue. The only way around this would be to ship the 6.0 GA version of PFC. But, of course, you can't do that because then you don't get bug fixes. Developers should do a full rebuild any time they upgrade to a newer version of PFC.

DESCRIPTION: GarbageCollectSetTimeLimit ( newtimeinmilliseconds ) Sets the minimum interval between garbage collection passes.
ARGUMENT: newtimeinmilliseconds - A long (in milliseconds) that you want to set as the minimum period between garbage collection cycles If NULL, the existing interval is not changed

Specifies the minimum interval between garbage collection passes: garbage collection passes will not happen before this interval has expired.
Garbage collection can effectively be disabled by setting the minimum limit to a very large number. If garbage collection is disabled, unused classes will not be flushed out of the class cache.

When I pass GarbageCollectSetTimeLimit a value equal too 2147483647 (maximum value for a long) there is no effect.

Recreated with 32 bit PowerBuilder 6.0 GA and 6.0.01 build 424.

Customer has created extended attributes in Powerdesigner and wants to use these in Powerbuilder. He has created public synonyms with these extended attributes. The public synonyms are available, including the extended attributes, while in PB in the Database Painter and when creating a Quick Select dw. They are not available in a SQL Select datawindow.

He is trying to create datawindows with these public snyonyms and he can see them when he chooses Quick Select, including all the extended attributes, but when he tries to build a SQL Select datawindow he can see a table reference with the same name but it appears without the public qualifier and it does not have any of the extended attributes. He needs to use SQL Select because several tables are usually involved in these datawindows.

This does not appear in 5.0.04 or 6.0.01.

You must have an intermediate structure that will take the return from the function and then assign the array element to a variable.

This was allowed in 5.0

For example:

uf_function() returns a structure of type 'LongStructure' that contains 2 elements: One of type Long, and another of type Unbounded Long Array.

i.e.

LongStructure

long l_long

long l_longarray[]

long ll_rtnlong

ll_rtnlong = uf_function().l_long // this will work in 5.0 and 6.0

ll_rtnlong = uf_function().l_longarray[1] // this will not work in 6.0, but works in 5.0

Both of these cases in the future will generate compiler warnings. Returning a structure as part of a function call in a Powerscript statement can be dangerous since the memory could have been freed. We made several changes in 6.x and beyond to fix several memory leaks that were really these structures left lying around (which is why you could reference them as above in 5.0). Now that we clean these structures up after the function call, it's possible to get memory corruption with these types of constructs.

Correct way to code this:

long ll_rtnlong

LongStructure ll_struct

ll_struct= uf_function()

ll_rtnlong=ll_struct.l_long

ll_rtnlong=ll_structl_longarray[1]

Per PB Engineering:

The reason this worked in 5.0 was because we were not deleting the temporary struct returned from the function call in the expression myfunc().fieldname. This resulted in a memory leak which has since been fixed.

As part of fixing this memory leak, the field elements were also free'd after the expression evaluation. Because the field here is a string, the string value therefore is free'd resulting in the garbage that you see when you run this test case.

It is expected that this syntax will work for simple data types at this time, but not for strings, or arrays.

Please issue a separate change request for a compiler warning against this, to eventually be a compiler error in the next major release of PB. Due to this, I would strongly recommend to customers that they change all occurrances of this syntax to use a local struct variable, and return the function result into that. This ensures that the data will be correct and avoid any crashes until we fix the compiler to warn against this.

(CR 203969 was added to request the compiler warning.)

All show the same 'weird' numbers on the axis like 49.9999999 & 149.9999997 instead of rounding to 50 or 150.

Workarounds:
1. Turn autoscale back ON in the Value Axis Tab Page.
2. Format the Value Axis Text by clicking on the Text tab of the graph property sheet.
-Click on 'Value Axis Text' item in the Text Object listbox to show all properties for the Value Axis
-Notice the Display Format field shows [general] for the format. Click on the Formats... button and change the format to something like 000 or ### and you will no longer see trailing 3's or 9's.

Through testing his original full application (44 pbls), and the narrowed down sample app I discovered that the Full Rebuild started failing in Build 404, which is the first available build after a different PFC application Full Rebuild (CR 164588) bug was fixed. This was tested up through build 444 with the same results.

The Setitem() call causes an allocation (to hold some data)

This was due to some extra source for some of these objects causing a problem when made global. To permanently correct the source, you need to resave all the inherited objects using the user object painter from pfc_n_base down to the object you're using as a global. Migration, regeneration and import will not solve the problem - the source must be re-written by the painter.

In general, we don't recommend using autoinstantiated objects as global variables.

Workaround: Code the references to these objects from the application manager NVO inherited from n_cst_appmanager (which of course is not autoinstantiated). This way, there is only one global variable - everything stays encapsulated, and it is not hard to reference in code (just qualify everything with gnv_app.reference)

The dll's being left in memory is happening on all platforms tested, WIN95, NT 4.0 and WIN 3.11 platforms. Extensive testing was done with WIN95 as this is the platform that this customer is using.

When the customer runs a 16 bit exe with datawindows or userobjects with datawindows on them PB is leaving 11 - 16 bit dll's in memory. When the customer then tries to run a 32 bit application he is receiving an illegal page fault. Also trying to rerun the 16 bit application or returning to the 32 development environment causes gpf.

After testing this issue out I found the problem to exhibit itself in different ways depending on what I tried to do next after closing out normally from a simple 16 bit executable. I also tried this with both machine code and pcode with the same results. The PB 6.0 exe leaves the following 11 dlls in memory: ole2.dll,ole2disp.dll, ole2nls.dll, pbbgr60w.dll, pbdwe60w.dll, pboui60w.dll, pbvbx60x.dll, pbvm60w.dll, storage.dll, typelib.dll, unidrv.dll. Furthermore, when I tried to unload the dlls if I unload pbdwe60w.dll all of the other 10 dll's also get unloaded but if I try to unload pbvm60w.dll then pbbgr60w.dll and pbdwe60w.dll were still left in memory.

Workaround:

Declare as external functions:
SUBROUTINE FreeLibrary(uint libhandle) LIBRARY "Krnl386.exe"
FUNCTION int GetModuleHandle(string modname) Library "Krnl386.exe"
Script in Close Event of Application:
uint l_handle
l_handle = GetModuleHandle("pbdwe60w.dll") // This can be any dll or library that is loaded into memory.
FreeLibrary(l_handle)

Fails with following 3D graph types :
3D Column
3D Bar
3D Area
3D Line

2D graph datawindows print fine and MS Graph prints correctly with a 3D graph.

Trying to delete or re-enter a new date on a highlighted editmask column with either Center or Right alignment that is only partially visible because the column is too narrow results in an illegal page fault or a seemingly infinite loop. Left (default) and justify alignment appear to not exhibit the problem.

With the infinite loop when I hit Ctrl=>Alt=>Delete I got a messagebox with "The system is dangerously low in resources. Would you like to terminate the following application? accompanied by the title of the window. "

Illegal page faults I received on my machine : PB 6.0 GA build 251 @ 0137:1113c9cd
PB 6.0.01 Build 444 @ 0137:1111ef1b

Workaround: You can use a date() by instead substituting "DATE(datetime_column)" for VALUE on the graph.

function for truncate or make a round time "TruncateTime( time value, integer mode ) " truncate on hours minutes and so on - by mode selection

This only happens if you code the Update() function with the datawindow in this state. This was caused by a fix to 157312 in 5.0.04.

Customer has simple genapp application in which he is initially changes the visible attribute to false for several menuitems: visible. He then changes this same properties to 'true' in script and the application halts with an illegal page fault in USER.exe @ 0011:00000971.

Placing a messagebox, a yield() or running through the debugger eliminates the illegal page fault.

Must exit PB and start again to open the window.

Same happens if use a custom visual uo instead of a window in example above

Works OK in 5.0.04 but not in 6.0 GA nor 6.0.01 build 444

But. Change the name of the tab control in the ancestor object again and at the same time add a couple of comments to a tabpage in the tab control and this is the error which appaers in the descendant when trying to open and save for a second time...

Object: u_descendant
tabpage_1 Class Declaration
(0001): Error C0001: Illegal data type: tab_1

The only way to get this change into the descendant is export, edit and import the descendant object.

This can be reproduced with any control embedded in the tab control - reproduced with tabpages, datawindow controls, command button.

The important combination is
Change name of parent control (tab_1) once, allow changes to be saved in descendant
Change parent control name again AND add some script to an embedded control (or tabpage).

If you highlight all the data by doubleclicking and click on the DELETE button and tab off, you get "item does not pass validation test"

Development: Not a bug. It works this way because a single decimal point is not a valid number. When first painted we will paint a single "." but once data is entered or the contents of the editmask control the data window engine will insist on a valid number being there. One workaround would be to define the edit mask as #.00, and define a format display of #.##, define the column to "Use Format". Then, when the column has focus, the column will display .00 but, when it does not have focus it will display "."

Setting QueryMode on without QuerySort is OK in the same conditions.

If you place a command button on a window which contained the following script: MessageBox("Today's Date",string(today())), when this is executed the messagebox will display today's date e.g. 06/22/98

If you edit the .WindU file and scroll to the [intl] section, you can see the date format which is used.

Here are the initial settings:

SShortDate=M/d/yy
SLongDate=dddd, MMMM dd, yyyy

If you now change the above to European date format i.e.

SShortDate=d/m/yy
SLongDate=dddd, MMM, yyyy

Now if you execute MessageBox("Today's Date",string(today())), this will return e.g. 22/06/98. Which is what you would expect.

If you now retrieve some information from your database via your datawindow, which contains a datetime column, the dates would be displayed in European format. Which is what you would expect.

If you try and enter a date using the European date format dd/mm/yy, in preview or runtime mode, PowerBuilder will give you the following DataWindow error message "item '29/03/98' does not pass validation test." If you typed in 03/29/98, then there is no problem, UNTIL you tabbed off the column. The date gets converted into European format i.e. 29/03/98.

This is not a regression.

When the user clicks on an actual column this does not happen. There also does not appear to be problem with the analogous "alway show arrow" property checked or unchecked.

char char5[5]

char5 = "1234567890" // will give -> 12345 Correct, expected behaviour
char5 = dw_1.object.col1[1] // will give 1234567890 (assuming the col has that value)

A workaround is to pass the same code through String()

Development: The problem is with the browser/window painter in that it builds a list of objects has an architectural limitation.

Development: Verified that leak is caused by Oracle client not giving up memory associated with the stored procedure calls until the environment handle is is freed by a call to OCIHandleFree.(). Developer needs to contact Oracle to open a case with them.

The settings that fail:

New Page on Group Break (ON)
Reset Page Number on Group Break (ON) <----- This is what is broken.

This allows things like

Page 1 of 2 - Page 1 for value 100
Page 2 of 2 - Page 2 for value 100

Page 1 of 3 - Page 1 for value 200
Page 2 of 3 - Page 2 for value 200
Page 3 of 3 - Page 3 for value 200

What happens now:

Page 1 of 2 - Page 1 for value 100
Page 2 of 2 - Page 2 for value 100

Page 3 of 3 - Page 1 for value 200
Page 4 of 3 - Page 2 for value 200
Page 5 of 3 - Page 3 for value 200

Workaround: Placing the datawindow in a composite will show the Page() correctly as the groups break and continue from page to page.

Put on a trace and you see that the table pbcatcol is being changed for this table/column for the column when you only change the justification attribute...

pbc_case from the value 27 (upper) to the value 0 (lower)

Would like this value not to be touched in PBCATCOL if it hasn't been changed in the painter

Not a PB bug. In 32 bit passing an address by value between two separate apps will not work as they are running in
separate memory spaces, and cannot access each other's memory.

Add code to a descendant event which has no code in the same event in the ancestor. Go back and add code to this ancestor event. When we return to the descendant event the new code in the ancestor has been OVERRIDDEN by default. In PB 5.0 this was EXTENDED by default.

Possible workaround: perform calculations on the back end if possible, and return as part of the result set. Then the calculated data
will be treated as regular columns by the nup datawindow, and SelectRow will work properly.

GetFileOpenName() and GetFileSaveName() have the same results at runtime. Once the working directory has changed, finding .ini files with ProfileString() calls often fail.

Workaround: Reset the working directory by opening (expanding) a PBL in the directory that you want to be your working directory.

Simple workaround is to use the IsDate() on all dates entered in SLE's to insure that they are valid.

In PB 5.0.04 after the initial PB Messagebox for the DataWindow error the application allowed the user to enter a correct value or drop down the DDDW and pick from the list.

Problem happens using using distributed pb and calling remote method that's retrieving a nested report.

Getting the following error on hpux:
The error of Floating exception pb60 (core dumped) will appear on the unix command line.

On solaris, the error is:
signal fault in critical section
signal number: 11, signal code 1, fault address: 0x0, pc: oxecfa35co, sp: 0xeb0058f8
ABORT: core dump

Testing:
Used apache web server on solaris and build 444 of pbcgi60.exe. App server was on hp and solaris. Also tried using website on nt 4.0 and app server running on hp and solaris. Same problem with pbcgi60.exe on windows.

Doesn't happen in Solaris.

When a field within a datawindow is too wide or extends too far to the right for the size of paper to be printed on, the datawindow displays a second empty page regardless of the content of this field at runtime. Microhelp at bottom of datawindow in preview mode says "Horiz page 1 of 2"

Instead of taking the runtime width and calculating when a second page is needed the datawindow seems to base its calculation of the static width of the field as defined in the datawindow painter so it shows an empty page when this is not necessary.

For example a string field has a max width of 50 chars, when there is a string longer than 25 the field will not fit on 1 page so it prints on 2. But when this same field only has 1 char it still shows a second page even though it is blank. Customer would like the datawindow not to show this page when it's not needed by checking the runtime width of the column/field and not width as drawn in the datawindow painter (which is always the same, in this case 50)

Composite Parent
Header report
Composite Report #2

The "Composite Report #2" would start immediately after the Header Report on the FIRST page. (assuming the header report was less than 1 page).

Now, in 6.5.00, when previewing or printing the report, if the entire "composite report #2" does not fit onto the remainder of the FIRST page, it is started on the SECOND page.

Workaround: Place the Parent composite report above into another composite report and by adding one level, it previews and prints as expected.

Changed the sShortDate in the Intl section. Same problem.

Ran a test using pb 5.0.04 on Solaris, no problem. I was able to select all 26 columns--no problem.
Under pb 6.0.01 (build 444), pb will core dump on solaris.

Ran another test on solaris (build 444) and selected only 20 columns. I then went back into the dw painter and selected
the additional 6 columns 1 by 1 and added them to the dw. No problem. I then hit design/preview
and the following error message appeared on the unix command line:

MEM_BAD_MEM_POOL
Abort, Ignore?
(I then hit Ignore and the following message appeared)
Premission denied

All of this works under pb 5.0.04 on Solaris.

This may be correct behavior since at this point it is pretty difficult to anticipate what the user's next move will be, much like the situation
where a spin control allows invalid dates, but user would like development determination.

PER DEVELOPMENT:
This works as designed. The editmask control does not check valididity when it loses focus. The DataWindow engine does check valididity when tabbing from one column to another which is why it fails there. Users should check themselves in the LoseFocus event.

instead of post open(w_response)

Per Development:
This is a problem with running the ue_accepttext event when the DataWindow control has focus. By running AcceptText(), the validation is triggered again, which pops up a message box, which causes the control to lose focus, which run the LoseFocus() event which posts another event. In 5.0, the ue_accepttext event ended up running while the message box was up, so the DW did not have focus. The fact that the event ran while the message box was up was a bug that was fixed in 6.0 (it was causing lots of problems). Coding the event in this way is incorrect for 6.0. A way of fixing it in the PowerScript would be to keep track of whether the DW has focus (using an instance variable and the GainFocus()/LoseFocus() events) and not running the AcceptText() in ue_accepttext if the DW has focus. I tried this solution in the supplied example and it fixed the problem.

Error message:
"Clearing check out status of library entry failed. Probably library I/O error."

The following script will produce the access violation:

lstr ls_mystructure[]
any la_any

la_any= ls_mystructure
cb_1.text = classname(la_any)

Workaround: Use 0.xx for data entry. That is, enter a zero, then a dot (or comma if your decimal separator is set up as such) then the decimal number.

Tested this with several types of windows with results as follows: Response windows - debugger comes up behind response window
Popup window - debugger comes up behind popup window
Main window - debugger comes up in front of main window.

Workaround: Make sure response window is small enough while in development to 'grab' debugger window behind it.

1. Retrieve (currentrow = 1 after retrieve)
2. Change the current row (click on row 2)
3. Retrieve again (currentrow changes to 0 instead of 1 in 6.0.00 and 6.5.00)

On Windows 95 and on NT4.0 the memleak_3.pbl test case is leaking about 4.2M of memory per 10K of test case loop iterations or about 420 bytes per loop on average. On the HPUX machine it is leaking at about the rate of 5.4M per 10K iterations or 540 bytes per loop on avarage.

On all three platforms I observed some very interesting behavior: When I first bring up PowerBuilder and start the test case, it starts leaking right away and continues to leak through out the first test run. For example on the HPUX, memory would initially be at 7.3M for PB06.exe and after 10,000 iterations the memory usage would 12.7M. If I then exited the test case execution, returned to the development environment, and then restarted the app, memory usage is still at 12.7M. But now when I run the app it does not leak until around the 10,000th iteration! Then it starts leaking again at the same rate as before. So it appears that in between executions the memory that was 'leaked' or used is reclaimed by PB. It does not return it to the operating system, but keeps it internally and is able to reuse it. But it is not able to reclaim this memory during the execution of a single test session, even when the app is idling. I can suspend the loop iteration and idle, but no memory is reclaimed. Resuming the test loop at this point continues the leak.

So knowing this, the best way to always see the leak is to shut down PowerBuilder and restart it, then the test case will leak from the start. Otherwise any given test case run is dependent on the previous highwater mark of memory usage in the same PB session before it will start to leak again.

I've simplified the test case and can also see the memory leak with a simple nested report created from the demo db. The test case has a main datawindow with 1 row of data, and a nested report with one row of data. No database connection is needed, the datawindows have data retained on save. The script is a tight loop that simply assigns a string to a column in the nested report using dot notaion. The leak is isloated to the line of script that is dot notating into the nested report and assigning a string to a column in the nested report:
.
dw_1.object.d_emp[1].object.emp_fname[1] = "ABCabcDEFdef"
.
If the line assigning the string to the dot notated column is commented out, it does not leak. The leak is related to the fact that the report is nested, if the nested report is put into the main datawindow control, dot notating to the same column no longer leaks. The size of the column does not seem to matter, very large columns and small columns all seem to leak. Also, the size of the string being assigned does not seem to be a factor, even assigning an empty string ("") leaks.
.
The problem is not related to visual processing, using a datastore leaks, and setting SetRedraw to false with a datawindow still leaks. The advantage of setting SetRedraw to false really speeds up the test case. It gets to the leak quickly and the leak is quite dramatic. There is often a delay for several seconds when starting the test (while the test loop is iterating) before the leak is apparent. I can only assume that PB has an internal memory pool and when the pool is exhausted it requests more free memory from the operating system. Once this initial threshold is reached, the leak is apparent, and rapid as PB keeps requesting memory and never returning any.
.
I also tested this new test case on both the Windows 95 and NT4.0 platforms. With this test case I'm also seeing a significant
leak on both of these platforms as well. It seems to be a smaller leak on these platforms, but it is clearly leaking.

Workaround: Turn off autosizeheight for both the nested reports in the header.

Address in PBOR860.DLL can vary a little.
0137:02026100
0137:02066100
0137:01ffc6100

The following SQL sentence causes a GPF

SELECT
nvl("TABLE_NAME"."COL_NAME",0)
+ nvl("TABLE_NAME"."COL_NAME",0)
+ nvl("TABLE_NAME"."COL_NAME",0)
+ nvl("TABLE_NAME"."COL_NAME",0)
+ nvl("TABLE_NAME"."COL_NAME",0)
FROM
"TABLE_NAME"

Most often happens with datawindows with groups in development and runtime but the printout is OK. Nothing seems to get rid of the blank areas during display (setredraw, etc).

Development: There is a gap at the end of the page of grouped reports, where the next row does not fit and is pushed to the next page. This gap is still visible as the display is scrolled -- but no rows are missing. However, we did make a change to the group display. If the display is scrolled, a page separator (single line) is drawn between the two pages. This is an attempt to show that the gap is intentional, and the space is supposed to be blank. This line is not printed.

Workaround:
cast the any varibale into the rigth datatype and it works. ex. SetDynamicParm(1,integer(my_any_variable))

For a PROTECTED function called from a descendant, the DYNAMIC keyword seems to cause a runtime error and treats these function calls as if the ancestor function was declared as PRIVATE.

This was probably caused by the fix to 163944.

R0065 Application terminated. Error: Dynamic function not found. Possible causes include:
pass by value/reference mismatch...

This problem does not exist in PowerBuilder 6.0

This will not be fixed:
Per Development: This has always been the behavior for PB and would be too risky for Development to attempt a code change at this point. If this was done, it would have the potential to break too many existing applications that depend on the current behavior.

Workaround: The decimal() function seems to work OK and might be an acceptable workaround in Powerscript (but not in the dw painter functions)

Resolution from development: This is sorting as it should ... In Win32 the compare that we're doing internally intentionally ignores hyphens (-) and apostrophes (') (this is called a word sort as opposed to a string sort). To do otherwise would have to be an enhancement .

Workaround: To achieve a sort taking the leading "-" or "+" into account add the column you would like to sort that includes the "-" hyphens twice - assume it's called col1. Doubleclick on the first instance of col1 in sort dialogue and modify the expression as follows ( left(col1,1)). Check off sort order and leave the second instance as simply the column name. This will sort on the first character first and then on the entire column resulting in a sort that takes into account the leading "-" or "+".

WORKAROUND: Either make the column larger (usually only wider is necessary) or the font smaller, so that all of the data can be
easily displayed during data entry.

NOTE: For customers wanting a User Guide, please see page x in the InfoMaker Getting Started book. As in 5.0, the User's Guide is not packaged with the product. It can be ordered by calling the number on page viii.

dw_1.object.dept_id.dddw.required='yes'
dw_1.object.dept_id.background.mode='0'
dw_1.object.dept_id.background.color=rgb(192,192,192)

You get a GPF in MSVCRT.DLL

Ex: Select emp_id dept_name from employee, department where employee.dept_id = department.dept_id

Should product column names of: 'employee_emp_id' and 'department_dept_name' exactly what you'd see if you created a datawindow in the painter and looked at the 'default' assigned column names in design mode.

Instead, it produces 'cemp_id' and 'cdept_name'. This was an old behavior with Oracle but now this has migrated to all drivers (in this case SYC). Even with using the correct USERID prefixing, there is no way at runtime or in development to get the correct names.

This is definitely related to CR 161648.

Workaround: Prefix the tables in the FROM clause with their owner.

Ex:
Select emp_id, dept_name from employee, department where employee.dept_id = department.dept_id
becomes
Select emp_id, dept_name from dba.employee, dba.department where employee.dept_id = department.dept_id

Per Development:
This will not be fixed, you must use the workaround.

String (real (sle_real.text))
An editmask on a datawindow column has the same behavior when using the Real() and Double() report functions in the datawindow as computed fields.

Workaround: Full Rebuild.

Workaround: Use a nested report instead of a second group on one report.
The parent report would define the first group (you might have to change the SQL statement and use DISTINCT).
The nested report would contain the original 2nd group but have this as it's first and only group.

An example can be given to all customers by using the PBL for this bug (it's much easier to see than to explain).

The problem with this workaround is it creates a nested report which will have the "1 page per detail row" limitation (and this could cause problems with some reports where the nest flows beyond one page)

NOTE: This has been fixed in 6.5 and above. Any build after 500 should work. For example 6.5.1 build 620 and above should work.

In the db are three columns of integer and one of type money. The three integer columns are the primary key.
Crosstab dw definition:
define the first integer = column, and the 2. and 3. integer are rows. Value is the sum of the money . Now in some special cases when the column (first integer) is equal to the 2. integer row the crosstab dw mixed the values of the integer rows.

Workaround:
Insert a computed column between the two rows and make it invisible

Strangely, the assignment of a structure element of type datetime to datawindow column date using dot notation works as does assigning a structure element of type date to a datawindow column of datetime.

Workaround: Use exactly matching datatypes.

This appears to be related to 161211 which was fixed in 6.5.00. The fix for that bug seems to have also fixed this particular offshoot.

Under Win95:
"PB60 caused an invalid page fault in module PBDWE60.DLL at 0157:115e83fa. "

Under WinNT:
"The instruction at 0x115e83fa referenced memory at 0xd0000005e. The memory could not be written"

Only occurs when performing the above with a standalone application - does not occur with a distributed application where the GetFullState() is done on the server and the SetFullState() is performed on the client.

HKEY_CURRENT_USER\Software\Powersoft\PowerBuilder\6.0\initpath

The PowerBuilder 6.5 environment seems to work OK. The profiler.exe seems to want to read from the '6.5' key:

HKEY_CURRENT_USER\Software\Powersoft\PowerBuilder\6.5\initpath

Change Request 171575 addresses that.

It almost seems like when the 64k limit is reached, Powerbuilder 'acts' as if the pb.ini is readonly, missing entirely, or the PBL that the DefApp key points to has been renamed or moved and puts up the New/Open/Cancel dialog for creating a new application.

Workaround: Edit your pb.ini and delete applications from the bottom up
Look in the [Application] section. The applications begin with a $. Delete until you're way under 64k and you will find Powerbuilder will automatically open up the correct 'DefApp' application on startup again.

PB 6.5 REGRESSION :

Following script causes a crash in 6.5, either immediately or after exiting the application

>>>

scriptdefinition scriptlist[]

variabledefinition argumentlist[]

..

argumentlist = ScriptList[119].ArgumentList

<<<

This problem is related to CR162727

>>>

Crash while accessing the properties of created Meta Information variables (EnumerationDefinition, VariableCardinalityDefinition, ScriptDefinition,

VariableDefinition)

<<<

When creating a new DW and hit 'X' to close, click cancel to save and the Preview Icon on toolbar is disabled. Design=>Preview works fine and also hitting SQL icon to get out of SQL mode icon is enabled. Even when disabled on machines clicking Icon once changes it from looking disabled to enabled and it worked properly.

WORKAROUND: Hit icon twice or exit by hitting SQL icon rather than "X".

What to do: Just delete the files manually from the directory where you directed the program to unpack them.

Another set of bad side effects
- any filter() or retrieve() directive seems to break the change. (reads in the original dataobject definition - this is incorrect)

examples of the dot notation:

dw_1.object.d_employee[1].object.emp_id.color="0~tif( emp_id < 200, 255,0)"
dw_1.object.d_employee[1].object.emp_id.font.weight="400~tif( emp_id < 200, 700,400)"

Workaround: Use Full Rebuild.

Workaround: Set myconnect.options = "MaxRetry=16" or anynumber > 16.

In fact in the Development environment, using LibraryImport() with the syntax from Describe() prevented the page faults with the previous two methods of generating the SQL even when the active pbl was changed. If the PB development environment was exited and re-entered the page faults reappeared.

Behavior was consistent with above in an executable version.

Workaround: Place a empty invisible DW control on window.

This is a case where the user wants to see the warnings or errors (IE they don't want to UNCHECK showing of Information or Obsolete Messages in the Design/Options menu of the Project Painter (7.0)) but they don't want them to be shown - perferably with an option to log to a file - until after the exe is created. (esp if the messages are just warnings)

If the project is run as an automated process (through a tool such as sqa robot), there is no way to continue on until a user actually responds to the messagebox which makes it hard to automate overnight builds and running of regression suites afterwards.

A batch mode compile would also be nice.

PER DEVELOPMENT:
(Nov. 17/98): this issue is dependent on the printer driver and its configuration settings; I printed the sample on a desktop Canon and on a HP 4Si LaserJet (using a PostScript driver); I received gray-scale print on the Canon and black print (on all columns) on the HP; if the printer can handle gray-scale printing then it will turn colors into shades of gray; if the printer does not handle gray, then it will round down the color value to a black, or round it up to a white (which makes it invisible); it could be a problem with PrinterControlLanguage (PCL) drivers vs. PostScript drivers; try altering the Graphics properties of the printer driver to produce darker, finer, vector output -- this should cause more colors to round down to black, instead of white

A better approach is to use GetFullState() and SetFullState() only when initailizing with data retrieved from the database and then use GetChanges() and SetChanges() to reflect any changes. This does result in the correct status flags and updates.

Workaround: Use Getchanges() and Setchanges() to retain status flags.

NB:
-----
Sample pbl must be run on pb5.0.03, then on pb5.0.04, 6.0.00 and 6.5.00

When a datawindow is in QueryMode and the first character entered for a column is an international accented letter the datawindow rejects it with the message "Specified criteria is invalid".

A workaround exists.

WORKAROUND: The edit limit can be explicitly set within the painter.

PER DEVELOPMENT: This will not be addressed. The workaround is simple, and other users may prefer the current behavior.

When a Crosstab datawindow has Stored Procedure source, if you select Design...Datasource from the DW painter menu, you get to the crosstab definition dialog. I couldn't find a way to access the selection of the stored procedure or the retrieve arguments for the
procedure. If you create a tabular datawindow from stored procedure source, it is possible to modify or add retrieve arguments
through the painter. If you create a tabular datawindow from a SQL select statement, it is possible to add or modify retrieve arguments,
but it is not possible when the datasource is a stored procedure.

ALLBASE takes you to "Modify PowerScript function"
AllowEdit property takes you to "SelectItem PowerScript function: Syntax 2 When you know the item number"
Alphabetical list of properties takes you to a page entitled that, but it is blank.
Batch Applications takes you to "PostEvent PowerScript function"
condensed mode takes you to PrintSend Powerscript Function

Workaround: Replace '~r~n' with char(10), the ascii equivalent , and it works as before.
Example: "SELECT emp_id, emp_fname +char(10) + emp_lname FROM employee". Tested in both SQL ANywhere and Sybase SQL Server. Concatenation characters may be different depending on database.

From Development: The change in the DataWindow engine was intentional in Release 5.0. Thus this is not a bug and will not be changed.

Workaround: Turn garbage collection off using GCTimeoutLimit

This would help ct's dws print correctly and avoid an inconsistent dw behaviour (see below)

Current dw column default background color is 536870912 - transparent. But ct has problems printing this value on some printers. However 553648127 (also transparent) prints correctly.

Also if we change the column foreground color (from black to navy for example) then the background color for same dw column is changed from 536870912 to 553648127 (both appear as transparent in dw painter)

Therefore ct would like the default color to be 553648127 given that this is what the background is changed to when the foreground color is changed. This would make the datawindow painter behaviour more consistant and resolve the ct's printing problem.

WORKAROUND: Check for NULL in code.

1) connect to database

2) Retrive to datastore

3) Getfullstate()

4) SetChanges()

5) Update()

6) Disconnect

7) Retrieve()

PowerBuilder will GPF at the PrintStart event if the document contains a BMP image. If the BMP is removed, then there are no problems.

I get the same GPF with every version of PB I use: Kernel32.dll at 0137 : bff72978

Workaround:
in the selectioncanging() event of the treeview you can specify:

if dw_1.accepttext()=-1 then
dw_1.setfocus ()
return 1 //to prevent the selection for changing
end if

The problem is that the format is not always taking effect. If you 'tab' from one field to the other, the format will be used by the item that just lost focus, but if you press the down or up arrow to go to the row below or above you, the editmask stays in effect.

Also, if you scroll the datawindow and it is redrawn, all of the formats take effect as needed, except in the column that has current focus. If this column has the same 'useformat' specified, the entire column will change to look like the editmask.

Workaround: In the customers application the editmask and the format only differed for NULL values. To workaround the problem we added an expression to the 'color' attribute of the column as follows. "if (isnull( clock_in_1 ) and (getrow() <> currentrow()), rgb(255,255,255), 0)"
Since the formats seem to work correctly when going along the same row we only needed to check when the row changed.