Hello, FPC developers' list.

Today I tried drawf on windows again and found some new bugs.

1) Gdb outpus the next strings for RTL units:

file="../objpas/classes/lists.inc",fullname="D:/programming/fpc/rtl/win32/../objpas/classes/lists.inc",line="600"

D:/programming/fpc/rtl/win32/../objpas/classes/lists.inc does not look good :)

2) <-exec-next> steps very strange in the procedures:

1. procedure OutputV(V: Variant);
2. begin
3.  WriteLn(V);
4. end;
5. var
6.  V: Variant;
7. begin
8.  V := 'Test';
9.  OutputV(V);
10. ...

- line 9: -exec-step
- line 3: -exec-next { Why line3 instead of line2? }
- line 4: -exec-next { correctly stepped to Line4 }
- line 3: -exec-next { Why we are back to Line3 ? }
- line 4: -exec-next { We are still inside the OuputV }
- line 10: { finally we can exit }

It is hard to debug something when stepping works non-reliable.

Stabs works more correctly here.

Paul Ishenin wrote

Hello, FPC developers' list. Today I tried drawf on windows again and found some new bugs.

Sorry. Forgot to attach a gdb session. It can be useful. http://www.hu.freepascal.org/fpcircbot/cgipastebin?msgid=18844

Best regards,
Paul Ishenin.

What's wrong with it ? it is a valid path and it points to the file

Marc

Marc Weustink wrote

> D:/programming/fpc/rtl/win32/../objpas/classes/lists.inc does not look > good :) What's wrong with it ? it is a valid path and it points to the file

My fault. Probably some lazarus bug prevents opening of the file.

Best regards,
Paul Ishenin.

Paul Ishenin wrote

It is hard to debug something when stepping works non-reliable.

Marc has exmplained me why I have the strange stepping. It is due to internal try finally which is generated for variants.

I think the solution is to set the same line number for that try-finally nodes as for the variant itself.

Or as Marc suggested - skip generation of line info for the internal try blocks.

Best regards,
Paul Ishenin.

I made more tests and Variants are not related to the problem.

Then next simple application also has problems with stepping:
program project1;
{$mode objfpc}{$H+}
uses
  SysUtils, Classes;
var
  List: TList;
begin
  // you need to step into 2 times to be in the TList constructor. In
the constructor you will visit first and last lines 2 times
  List := TList.Create;
  List.Free;
end.

Best regards,
Paul Ishenin.

A constructor also has an implicit try/finally, so the fact that you visit the first and last lines twice is probably the same thing as with the variant. I cannot reproduce the need to "step into" twice on either Mac OS X/i386 or Linux/i386.

Jonas

Jonas Maebe wrote

A constructor also has an implicit try/finally, so the fact that you visit the first and last lines twice is probably the same thing as with the variant. I cannot reproduce the need to "step into" twice on either Mac OS X/i386 or Linux/i386.

Probably this is something else. Maybe win32 only as I visit first and last for any routine I checked.

Best regards,
Paul Ishenin.

This should be fixed now (for all try/except and try/finally blocks).

Jonas

Jonas Maebe wrote

This should be fixed now (for all try/except and try/finally blocks).

Thanks, fix confirmed.

Best regards,
Paul Ishenin.