heaptrc - 3 time false positive

[winni]

Hi!

I ported a project from laz 1.6 to laz 2.0. In 1.6 there where no memory leaks - if you can believe heaptrace. In 2.0 suddenly 3  not  freed blocks appeared - pointing to strange lines of code

1. Pointed to rasterimage.inc, line 277, that is

Code: Pascal  [Select][+][-]

1. constructor TRasterImage.Create;
2. ...
3. -->     FSharedImage := GetSharedImageClass.Create;  

2. Pointed to a own function like

Code: Pascal  [Select][+][-]

1. function createSomeBitmap: TBGRABitmap;
2. var tmp : TBGRABitmap;
3. begin
4. ...
5. -->  result := tmp;
6. end;
7.  

3. Pointed to lpr

Code: Pascal  [Select][+][-]

1. --> Application.CreateForm(TForm1, Form1);
2.  

These infos even apeared after "clean up and compile"

That all seemed to be nonsense.  I activated  valgrind in the debugger  options to see  if I could find some errors. After compiling again and looking into the heaptrc.log, all 3 errors where gone.

Spooky, spooky.

Anybody else with similar experience???

 Lazarus 2.0, fpc 3.04, Suse Tumbleweed

[jamie]

I think you'll find the bug is most likely still there, just covered up with the difference of code generation that
takes place with the option you selected, but then again I could be wrong.....


 Recently just prior to 2.0.2 release I found a bug and others helped to locate it and come up a plan to fix it..

The bug is in the TRawImage.LoadFromRawImage call, it actually recreates more chucks of memory when copying the
image and not freeing the original..
 
 Maybe what you are seeing is something in that component getting effective by it?
Put this in the interface section
and this in the

Code: Pascal  [Select][+][-]

1. TBitmap = Class(Graphics.TBitmap)
2.  procedure LoadFromRawImage(const AImage: TRawImage; ADataOwner: Boolean);
3. end;
4.  

implementation sectect.

Code: Pascal  [Select][+][-]

1. procedure TBitmap.LoadFromRawImage(const AImage: TRawImage; ADataOwner: Boolean);
2. var
3.   img: PRawImage;
4. begin
5.   BeginUpdate;
6.   try
7.     Clear;
8.     if AImage.Description.Format = ricfNone then Exit; // empty image
9.  
10.     img := GetRawImagePtr;
11.     img^.Description := AImage.Description;
12.     if ADataOwner
13.     then begin
14.       img^.DataSize := AImage.DataSize;
15.       img^.Data := AImage.Data;
16.       img^.MaskSize := AImage.MaskSize;
17.       img^.Mask := AImage.Mask;
18.       img^.PaletteSize := AImage.PaletteSize;
19.       img^.Palette := AImage.Palette;
20.     end
21.     else begin
22.       // copy needed
23.       img^.DataSize := AImage.DataSize;
24.       if img^.DataSize > 0
25.       then begin
26.         //GetMem(img^.Data, img^.DataSize);
27.         ReallocMem(img^.Data, img^.DataSize);
28.         Move(AImage.Data^, img^.Data^, img^.DataSize);
29.       end
30.       else img^.Data := nil;
31.  
32.       img^.MaskSize := AImage.MaskSize;
33.       if img^.MaskSize > 0
34.       then begin
35.         //GetMem(img^.Mask, img^.MaskSize);
36.         Reallocmem(img^.Mask, img^.MaskSize);
37.         Move(AImage.Mask^, img^.Mask^, img^.MaskSize);
38.       end
39.       else img^.Mask := nil;
40.  
41.       img^.PaletteSize := AImage.PaletteSize;
42.       if img^.PaletteSize > 0
43.       then begin
44.        // GetMem(img^.Palette, img^.PaletteSize);
45.         ReallocMem(img^.Palette, img^.PaletteSize);
46.         Move(AImage.Palette^, img^.Palette^, img^.PaletteSize);
47.       end
48.       else img^.Palette := nil;
49.     end;
50.   finally
51.     EndUpdate;
52.   end;
53.  

I put that in the unit where I was using RawImages.LoadfromRawImage and it fixed the leak

the commented lines are the original, the allocatmem is the added lines.

P.S.
 I guess I didn't submit soon enough, this bug is still in 2.0.2

[winni]

Hi!

Thank you Jamie for diving deep in the rasterimage.

The strange thing is: after activating the valgrind option once, all the  leaks are gone, even if I disable valgrind again. So there are no memory leaks anymore. And I never found those kind of leaks in 1.6.

Perhaps this Information helps you.

Winni

[Thaddy]

Plz note that your original code did NOT generate false positives: it leaked! although your code is correct.
The cause of the leak is now fixed.

This may not be immediately obvious to other readers.