Messages

1

Graphics / Re: crtdbg dumps a memory leak when sf::Text::setOutlineThickness is used

« on: June 19, 2017, 05:55:05 am »

I'm not certain how FT works in that case, but it seems to me, that the glyph may get replaced with the "stroked" glyph. But by not destroying the origin/source glyph, SFML creates a memory leak, as the clean up code only destroys the "stroked" glyph and never destroys the source glyph.

Exactly this happens.

FT_Glyph_Stroke(...) definition is in the link in previous message.
All functions except FT_ALLOC(...) are defined in
http://git.savannah.gnu.org/cgit/freetype/freetype2.git/tree/src/base/ftglyph.c

FT_Glyph is defined as a pointer to a struct FT_GlyphRec, which is defined in
http://git.savannah.gnu.org/cgit/freetype/freetype2.git/tree/include/freetype/ftglyph.h.

Inside FT_Glyph_Stroke(...) a new pointer FT_Glyph is created and origin glyph is copied in the new FT_Glyph using FT_Glyph_Copy(...) function, which allocates memory to the new glyph, so the new FT_Glyph now points to memory containing the copy of origin.

FT_EXPORT_DEF( FT_Error )
  FT_Glyph_Stroke( FT_Glyph    *pglyph,
                   FT_Stroker   stroker,
                   FT_Bool      destroy )
  {
    [...]
    FT_Glyph  glyph = NULL;

    [...]

    glyph = *pglyph;
    [...]
    {
      FT_Glyph  copy;

      error = FT_Glyph_Copy( glyph, &copy );   // Memory to origin glyph copy is allocated here
      if ( error )
        goto Exit;

      glyph = copy;
    }
    [...]
 

FT_Glyph_Copy(...) calls ft_new_glyph(...), which in turn calls FT_ALLOC(...), which allocates memory for the copy.

After manipulations with copy (outline and other things related to modification of the glyph) pointer *pglyph is redirected to the new FT_Glyph, and if origin glyph isn't destroyed by specifying destroy argument as "true" (which would call FT_Done_Glyph(...) on origin), memory for origin glyph is left hanging, which supposedly produces the leak, because origin glyph isn't used anywhere in subsequent code in Font.cpp.

   
[...]
if ( destroy )
      FT_Done_Glyph( *pglyph );   // Memory for origin glyph is cleared

    *pglyph = glyph;    // If destroy is "false", origin glyph wouldn't be destroyed and will be left hanging
                        // for someone else to use/clear
    goto Exit;

[...]
} // the function end
 

So it seems to me that setting destroy argument to "true" is a fix.