Bug 101866

Use cairo_raster_source_pattern_set_callback_data() to set the closure argument of your callback: https://www.cairographics.org/manual/cairo-Raster-Sources.html#cairo-raster-source-pattern-set-callback-data

The pattern argument is a static, stack-allocated copy that cairo creates internally. I have to admit that I am not sure why this pattern is given to the callback at all. The backtrace for the creation of the copy looks like that (I guess that both calls create copies of the raster pattern):

Breakpoint 2, _cairo_pattern_init_static_copy (pattern=pattern@entry=0x7fffffffe690, other=other@entry=0x55555575aab0) at cairo-pattern.c:368
368	{
(gdb) bt
#0  _cairo_pattern_init_static_copy (pattern=pattern@entry=0x7fffffffe690, other=other@entry=0x55555575aab0) at cairo-pattern.c:368
#1  0x00007ffff7aa7552 in _cairo_gstate_copy_pattern (original=0x55555575aab0, pattern=0x7fffffffe690) at cairo-gstate.c:937
#2  _cairo_gstate_copy_transformed_pattern (pattern=pattern@entry=0x7fffffffe690, original=0x55555575aab0, ctm_inverse=ctm_inverse@entry=0x55555575a690, gstate=<optimized out>)
    at cairo-gstate.c:946
#3  0x00007ffff7aa9346 in _cairo_gstate_copy_transformed_source (pattern=0x7fffffffe690, gstate=0x55555575a540) at cairo-gstate.c:973
#4  _cairo_gstate_paint (gstate=0x55555575a540) at cairo-gstate.c:1063
#5  0x00007ffff7a978ca in INT_cairo_paint (cr=0x55555575a510) at cairo.c:2221
#6  0x0000555555554d9f in draw (cr=0x55555575a510) at test.c:50
#7  0x0000555555554dea in main () at test.c:65
(gdb) cont
Continuing.

Breakpoint 2, _cairo_pattern_init_static_copy (pattern=pattern@entry=0x7fffffffe3d0, other=other@entry=0x7fffffffe690) at cairo-pattern.c:368
368	{
(gdb) bt
#0  _cairo_pattern_init_static_copy (pattern=pattern@entry=0x7fffffffe3d0, other=other@entry=0x7fffffffe690) at cairo-pattern.c:368
#1  0x00007ffff7a9e59f in _cairo_composite_reduce_pattern (dst=0x7fffffffe3d0, src=0x7fffffffe690) at cairo-composite-rectangles.c:56
#2  _cairo_composite_rectangles_init (clip=0x0, source=0x7fffffffe690, op=<optimized out>, surface=0x55555575a380, surface@entry=0x7fffffffe690, extents=0x7fffffffe350, 
    extents@entry=0x7fffffffe2f0) at cairo-composite-rectangles.c:97
#3  _cairo_composite_rectangles_init_for_paint (extents=extents@entry=0x7fffffffe350, surface=surface@entry=0x55555575a380, op=<optimized out>, source=0x7fffffffe690, clip=0x0)
    at cairo-composite-rectangles.c:122
#4  0x00007ffff7a9f8ae in _cairo_compositor_paint (compositor=0x7ffff7dd7320 <spans>, surface=0x55555575a380, op=<optimized out>, source=<optimized out>, clip=<optimized out>)
    at cairo-compositor.c:55
#5  0x00007ffff7b0e391 in _cairo_surface_paint (surface=0x55555575a380, op=op@entry=CAIRO_OPERATOR_OVER, source=0x7fffffffe690, clip=0x0) at cairo-surface.c:2120
#6  0x00007ffff7aa9294 in _cairo_gstate_paint (gstate=0x55555575a540) at cairo-gstate.c:1067
#7  0x00007ffff7a978ca in INT_cairo_paint (cr=0x55555575a510) at cairo.c:2221
#8  0x0000555555554d9f in draw (cr=0x55555575a510) at test.c:50
#9  0x0000555555554dea in main () at test.c:65


For this bug report: I would argue that your expected behavior just is not the behavior that cairo provides, so there is no bug here. I'm not sure what you mean exactly with "it is dead, so doing anything with it crashes". Your example program does not crash here.

(In reply to Uli Schlachter from comment #1)
> Use cairo_raster_source_pattern_set_callback_data() to set the closure
> argument of your callback:
> https://www.cairographics.org/manual/cairo-Raster-Sources.html#cairo-raster-
> source-pattern-set-callback-data

Thanks. I'm using that now. I'd would have preferred to use cairo_pattern_get/set_user_data() so there can be no conflict if Python gets passed an existing object with configured callbacks/callback data from C. But I guess that's unlikely to happen anyway.

> The pattern argument is a static, stack-allocated copy that cairo creates
> internally. I have to admit that I am not sure why this pattern is given to
> the callback at all. The backtrace for the creation of the copy looks like
> that (I guess that both calls create copies of the raster pattern):

If there is no use for it I'll just not pass it to the Python callback then, thanks.

> For this bug report: I would argue that your expected behavior just is not
> the behavior that cairo provides, so there is no bug here. I'm not sure what
> you mean exactly with "it is dead, so doing anything with it crashes". Your
> example program does not crash here.

I meant that the object has a reference count of 0 and wrapping it, which involved refing/unrefing it, made it crash if I remember correctly. The example program should not crash but show a reference count of 0.

Closing is fine with me, thanks.

fyi, this is the Python API now: https://pycairo.readthedocs.io/en/latest/reference/patterns.html#cairo.RasterSourcePattern

Closing as per comment #2