Memory leak in FillText

[jimlears]

(commented on issue #135 in reference to the Context2d leaks that are also in here)

canvas = new Canvas(500, 500); 
ctx = canvas.getContext('2d');
ctx.textAlign = "center";
ctx.textBaseline = "middle";
ctx.fillStyle = '#ebebeb';
ctx.font = "32px 'Gotham Black'";
ctx.shadowColor = "#333";
ctx.shadowColor = "#252525";
ctx.shadowOffsetX = 0.1;
ctx.shadowOffsetY = -1;
ctx.shadowBlur = 0;
ctx.fillText("hello world", 10, 10);

[jimlears]

Follow up: Appears to be leaks all over the place.

[Tharit]

Regarding your first screenshot, those are the two leaks I fixed in my patch.

However, the problem in Context2d::SetFont seems to be related to the internals of 'cairo_select_font_face'. ]
As the documentation explicitly declares this as a "toy" function, it might be neccessary to switch to using a "real" font backend like cairo-ft. I'll try to investigate this.

[tj]

The toy function worked out quite well for what we needed, I'm assuming toy doesn't mean it'll have memory leaks, that would be a little odd but that might be the case

[jimlears]

It seems -- I am not 100% sure, though -- that the fix that releases that array of states in the Context2d destructor and called cairo_path_destroy() in Context2d::restorePath() has also made the (seemingly) SetFont leaks go away. Not getting those leaks in Instruments any longer.

[Tharit]

I can still observe that leak with my patches - however, it appears that not every call to SetFont leaks...

[Tharit]

It appears that the leak in FcPatternObjectInsertElt might be incorrectly reported by valgrind. It's suppressed in tests for other applications like for example gtk and chrome.

See:
http://src.chromium.org/svn/trunk/src/tools/valgrind/memcheck/suppressions.txt
https://bugs.freedesktop.org/show_bug.cgi?id=8428

Additionally, I have been running a node server serving several hundred images per hour for the last 4 days, and memory usage is not increasing (at least not noticeably).