I'm in an uncomfortable situation where the last few weeks' worth of trunk won't display a project's main form (populated only with standard controls) and won't compile it to anything reliable (noting other discussion, it's got a background thread). I got here because of fixes to the SPARC code generation which propelled me onto FPC 2.7.1, and fixes to Lazarus's gdb interface. After investigation yesterday I'm fairly confident that this problem is not CPU-specific, the first part (form display) appears to be something to do with GTK2.

I've got two questions:

i) If anybody thinks this is worth raising as a bug, what is the maximum amount of the project that I can attach as an example? I'm not confident that I can remove stuff from it using any Lazarus that can read and display it properly, but I presume that nobody'd want a project containing a dozen or so files.

ii) Allowing that the GUI-oriented stuff hasn't changed significantly in the program since it was on 0.9.30, how can I take the project and force it to be saved without any recently-introduced properties which now make reading unreliable on anything older than trunk?

What lazarus svn revision did you use ?
You are talking about gtk2 under X11 + sparc processor ? zeljko
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus.freepascal.org/pipermail/lazarus/attachments/20120211/1a86e1f8/attachment.html>

Can you insert debug code, to write the form coordinates, Visible etc. information into the console (or log file)?

DoDi

Sorry, I think I might have been ambiguous. It's the IDE which won't display the form when it tries to load the project, so it sounds suspiciously as though I might have to instrument IDE code- and quite frankly I don't know where to start.

When starting lazarus from console, do you get any error messages in the console?

Ludo

Nothing unexpected :-(

0 1>markMLl at pye-dev-07:/usr/local/share/lazarus-trunk$ ./*gtk2 Xlib: extension "RANDR" missing on display "localhost:11.0". TMainIDE.ParseCmdLineOptions:
PrimaryConfigPath="/home/markMLl/.lazarus"
SecondaryConfigPath="/etc/lazarus"
NOTE: codetools config file not found - using defaults NOTE: help options config file not found - using defaults TMainIDE.DoOpenProjectFile select form in designer: HeavyWethersForm:THeavyWethersForm TDesigner

So there is no problem with components not being created because of properties not understood, etc.

Could you take a look at the main form's lfm to check if the form coordinates (top, left, width, height) are well inside the screen. I don't remember the exact circumstances but there was in the past a problem with coordinates moving out every time the form was saved. Your form could simply be off screen or 1 pixel heigh.

Ludo

I think I've been ambiguous again. Running trunk the form displays near the top-left of the screen but only postage stamp size. Portions of the internal visible controls are visible, but again they're smaller than the form and are displayed near its top left (i.e. the top left of the form doesn't look too bad, but most of it is empty).

If I use the trunk IDE built for Qt rather than GTK2 then the form displays (some colours might be off), the project will compile but won't run reliably (thread issues?). To emphasise this, this is with all project files identical.

If I stop the trunk IDE, delete ~/.lazarus and start fixed (0.9.30.3) instead then the form is fully visible near the middle of the screen. Again, this is with identical files.

There are issues related to building the project using fixed, but I can probably revert to an earlier version of the code from my local svn i.e. back out of any UI "pollution" caused by trunk.

What thread issues you expect with Qt ? Maybe Qt lib have some problems under sparc.

Ok, there might be some change which isn't good for sparc. Can you find revision which broken things ?

zeljko
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus.freepascal.org/pipermail/lazarus/attachments/20120211/24282631/attachment.html>

I don't expect any problems at all: this code was working until a few weeks ago. However I note Henry's "Blank window" thread starting on the 26th January.

Maybe Qt lib have some problems under sparc. Ok, there might be some change which isn't good for sparc. Can you find revision which broken things ?

This is not SPARC: it's bog-standard x86 on Linux (Debian "Squeeze").

That was misleading, see the response :-(

Of interest are eventual error messages occuring when the application is started in the IDE (F9).

DoDi

Put an non-existing component property into a lfm and watch the console when you load the form ;)
You get also messages in the IDE but it wasn't clear from the original message what and if something was visible in the IDE. The console is also more verbose.

Ludo

This could be misleading, since one of the first things the program does is hide the main form (i.e. the one that's malformed in the IDE) and display a database login form in its place.

When I compile the program for GTK2 and start it (F9), I get what looks like a konsole window appear with

Xlib: extension "RANDR" missing on display "localhost:11.0".

While I believe the message is benign, I don't know what's opening the konsole session- I've not seen it in the past.

I also get an "Execution stopped" dialog(ue), plus an active database login window (a custom form, not a built-in).

On the konsole session used to start the IDE,

..
Compiling resource HeavyWethers.or
Linking HeavyWethers-i386-linux-gtk2
/usr/bin/ld: warning: link.res contains output sections; did you forget -T? 8281 lines compiled, 5.6 sec
23 warning(s) issued
43 hint(s) issued
12 note(s) issued
[TCompiler.Compile] end
[TMainIDE.DoRunProject] A
TMainIDE.DoInitProjectRun A True 0
TMainIDE.DoInitProjectRun B
TMainIDE.DoInitProjectRun
ProgramFilename=/usr/local/src/heavywethers/trunk/HeavyWethers-i386-linux-gtk2 [TDebugger.SetFileName]
"/usr/local/src/heavywethers/trunk/HeavyWethers-i386-linux-gtk2" [TMainIDE.DoRunProject] B
[TMainIDE.DoRunProject] END
PR: /usr/local/src/heavywethers/trunk/HeavyWethers-i386-linux-gtk2 --debug konsole(28994)/kdeui (kdelibs): Attempt to use QAction "change-profile" with KXMLGUIFactory!
[TDebugger.SetFileName] ""

If I fill in the database login form execution appears to continue, despite the "Execution stopped" dialog(ue), with a form displayed which should display data derived from a database on a server. However nothing useful is displayed, and a diagnostic (coloured asterisk) suggests that the background thread that talks to the database has done what's necessary but then things have ground to a halt.

So in short, I think there are (at least) two separate problems here: the first is an IDE problem affecting GTK2, and the second is a possible thread/sync problem which affects both GTK2 and Qt.

I know it's unwise to say this sort of thing, but I have no reason to believe that the application code is faulty: it's been running fine for more than a year.

hm...very strange. Can you post:
1.X11 version.
2.Gtk2 lib version
3.Qt lib version
4.kernel and libc version.

zeljko
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus.freepascal.org/pipermail/lazarus/attachments/20120211/4b64ab6f/attachment.html>

Reading between the lines, it sounds as though I'd be best going through all projects and making sure that they build and run properly with 0.9.30 before doing anything else. Thank heavens I set up svn here a few months ago.

If that is the route to follow and knowing that 0.9.30 warns about unknown properties, would I be better working from my most recent revision of each project and manually removing properties (from which files- .lfm?) or would I be better going to an older revision to get safe forms and then integrating more recent code?

Don't know when the functionality was introduced, but when you load a form using unknown properties you get a screen highlighting the problems and proposing to remove all unknown properties. If your 0.9.30 has already this functionality then going this route is clearly the easiest.

Ludo

Thanks Ludo, I've not seen that on either 0.9.30 or trunk... it sounds very much as though I've got a batch of screwed projects, due to moving onto trunk when I was having to use 2.7.1 because of SPARC problems. I think I should probably have a local policy of never committing a project if it's been saved by trunk.

I hate to admit this, but could you remind me of how to get library and X versions? I think you probably do it rather more often than I do :-)

In actual fact it happens on at least two systems: a SPARC based on Debian "Etch" and an x86 based on Debian "Lenny". And it didn't used to happen on either, since I was routinely building on both.

I'm planning to assume that I've got some screwed projects, focus on x86 (so as not to risk any bugs due to endianness, although I believe SPARC is OK by now) and go back in local svn to a point that will allow me to build reliably on 0.9.30+2.6.0.

I hate to admit this, but could you remind me of how to get library and X versions? I think you probably do it rather more often than I do :-)

'dpkg --list'. Be prepared for a lot of output :)

Ludo

There must be apt-cache or similar to get installed packages. I'm on rpm so rpmquery -qa | grep x11 lists all packages which have x11 in it's name , so libX11 too :)

zeljko
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lazarus.freepascal.org/pipermail/lazarus/attachments/20120211/85d96b36/attachment.html>

[Chuckle] although I'm sure that he'd have preferred the upstream version numbers rather than what Debian calls them.

Noting that X is going over ssh, on the system running the IDE and program (i.e. strictly the client)

xorg 1:7.5+8+squeeze1

and on the desktop system

xorg 7.1.0-19

Remainder on the system running the program:

2.Gtk2 lib version

libgtk2.0-0 2.20.1-2

3.Qt lib version

libqt4pas5 2.1Qt4.5.3-1

libqtcore4 4:4.6.3-4+squeeze1

4.kernel and libc version.

libc6 2.11.3-2

libc6-dev 2.11.3-2

Linux pye-dev-07 2.6.38-custom #2 SMP Thu Jun 9 11:07:18 GMT 2011 i686 GNU/Linux

You are aware that Mark's problem is loading the form in the IDE's designer?

Regards,
Sven

I have to take responsibility if he didn't, I'm not sure my first few attempts at explaining the problem were particularly rigorous.

Anyway, I've now reverted to 0.9.20.3+2.6.0, erased the project directory, reloaded from svn and found that Lazarus loads and compiles it happily. But a run throws lots of runtime errors similar to

TApplication.HandleException Error reading StringGrid1.Columns: Error reading TGridColumn.Title.PrefixOption: Unknown property: "PrefixOption"

Removing the affected lines from the .lfm gives me something that appears to compile and run OK. I'm going to work through any other projects I've touched over the last few months checking they're OK with 0.9.30 on x86 before doing anything else, but will revisit trunk presently.

I've managed to strip the project that started all this down, such that it's OK on 0.9.30 but fails to load correctly into trunk with no useful diagnostics. I don't know when the problem crept in, possibly around January.

http://mantis.freepascal.org/view.php?id=21792