Maybe you should start thinking about integrating a debugger in lazarus?

Anyone know of a location that has a good description of the available functions and the process of using libgdb??? I can look at the source but I would like to find some docs too.

Shane

That's great news, I'm looking forward to the nightly builds. Maybe I'll even be able to contribute!

Chris

you wrote

Anyone know of a location that has a good description of the available functions and the process of using libgdb??? I can look at the source but I would like to find some docs too.

I don't use libgdb, I use gdb as an external process. It is a bit more stable.

Marc

you wrote

I finally found the bug that was causing the Win32 SpeedButton problem. It was actually a Lazarus bug and had nothing to do with GTK. In fact I don't know how the Linux version was working correctly with the bug.

Wow! This is really something. Can you reveal what you found? I supposes it MUST have had some sort of an impact on the Linux side as well.

What about the severe memory leak I found? Did the bug have anything to do with it?

Markku

Yes and no:

- You can't derive a class from an object or vice versa.
- You can use the objects independently.

Using classes and objects intermixed is not really a beautiful programming style, though, also it is bound to producs hard-to-understand code.

Michael.

Congratulations, Michael.

I am curious though, what the problem was, I will be checking your patches carefully the next days.

Vincent

Florian Klaempfl wrote

Mixing both OOP models isn't a problem but e.g. streaming is a problem.

Okay, this is obviously correct.

TBreakpoint depends on the TV streaming. This is necessary because the breakpoints are stored in the desktop file. But using TV streaming in Lazarus is nonsense.

Yes

- Sebastian

Mixing both OOP models isn't a problem but e.g. streaming is a problem.

From ide\text\fpdebug.pas:

  PBreakpoint=^TBreakpoint;
  TBreakpoint=object(TObject)
     typ  : BreakpointType;
     state : BreakpointState;
     owner : PBreakpointCollection;
     Name : PString;  { either function name or expr to watch }
     FileName : PString;
     OldValue,CurrentValue : Pstring;
     Line : Longint; { only used for bt_file_line type }
     Conditions : PString; { conditions relative to that breakpoint }
     IgnoreCount : Longint; { how many counts should be ignored }
     Commands : pchar; { commands that should be executed on breakpoint }
     GDBIndex : longint;
     GDBState : BreakpointState;
     constructor Init_function(Const AFunc : String);
     constructor Init_Empty;
     constructor Init_file_line(AFile : String; ALine : longint);
     constructor Init_type(atyp : BreakpointType;Const AnExpr : String);
     constructor Load(var S: TStream);
                         ^
TBreakpoint depends on the TV streaming. This is necessary because
the breakpoints are stored in the desktop file. But using TV streaming
in Lazarus is nonsense.

you wrote

Have you , as Peter suggested, to see how it does it? Does it use libgdb? Or does it do something else? It seems to me that if the FPC guys have already figured it out, why not capitalize on that and reuse (or steal code) where possible.

Even more, I think we should think about a common interface to libgdb with some helper classes, i.e. that the non visual part is common for all IDEs. I know the classes/objects thing is a problem but maybe we can solve it.

BTW: WHich debugger are you using on Linux?

I use gdb when I do use a debugger. For this fix though I didn't use a was simply a stare at the code and trace through
it by hand solution. Anyway I was working on Windows to figure it out and trace it down.

Florian Klaempfl wrote

Even more, I think we should think about a common interface to libgdb with some helper classes, i.e. that the non visual part is common for all IDEs.

Yes

I know the classes/objects thing is a problem but maybe we can solve it.

We know about the class/object issue for almost 2 years, but I still don't know why it is a problem at all? Is mixing both OOP models in a single app really a problem?

- Sebastian

I finally found the bug that was causing the Win32 SpeedButton problem. It was actually a Lazarus bug and had nothing to do with GTK. In fact I don't know how the Linux version was working correctly with the bug.

I have an issue I want to deal with that involves the default font used in Win32 first before committing the code.

We should have usable nightly builds of Win32 Lazarus IDE by tomorrow.

Congratulations Michael!

Think I ran into that problem when I made the ShowMessage-functions. There was something wrong with the drawing of pixmaps in Linux but I decided to use a workaround.

BTW: WHich debugger are you using on Linux? I'm not satisfied with ddd because it sometimes locks up my system. There's a nice debugger from the gtkAda project but that one lacks pascal support and I can't compile it to introduce the missing module

bye,

Stefan

They are empty in cvs... yes. I already can start an app in debug :) I haven't done much lately, but someone can add the calls to the IDE,.

Marc

The FPC team uses gdb with patched Pascal support and/or the Linux IDE which has the same. (though the IDE there is still somewhat unstable).

Same for FreeBSD. (The status of Lazarus on FreeBSD is the same as on Linux, the FreeBSD 1.0.5 snapshot in the pub/fpc/snapshot directory on ftp.freepascal.org will probably compile Lazarus fine)

We could convert the object's in fpdebug to TObject's. Then we could inherit from them.

Shane

Is the plan to interface with gdblib or are we planning on writing our own? I may start looking into this if I can figure it out.

Have a look at the fpdebug.pas from the textmode IDE for all the debugger stuff.

Peter

Have you , as Peter suggested, to see how it does
it? Does it use libgdb? Or does it do something else? It seems to me that if the FPC guys have already figured it out, why not capitalize on that and reuse (or steal code) where possible.

Florian Klaempfl wrote

Maybe you should start thinking about integrating a debugger in lazarus?

Actually Marc has started that. He has the shell functions setup for it ... they're empty :-) but they are setup for it.

Is the plan to interface with gdblib or are we planning on writing our own? I may start looking into this if I can figure it out.

Shane

Actually it was a very simple mistake of using the wrong 2 variables. StretchBlt was using the destination pixmaps height and width for doing the copy instead of the source height and width. When the source pixmap was smaller then the speedbutton pixmap who knows what it was trying to copy and to where. That is why I am surpised that Linux didn't have a problem with it. There was also a problem with the size of the source glyph rect used when only one glyph existed but that wasn't as much of a problem.