Bug 72694

Created attachment 116405 [details] test modified to debug why length has such a big value (or negative)

I've spent a bit of time on this, so I'll share my findings - but I didn't find a solution yet. Timeouts on this test happens because the length of an image data is returning a very large number (random)... when it returns a negative value (I suppose it is because of some overflow), then the for inside compareRows that will check every pixel at the image will just not run. Now trying to understand why this length is getting this strange value, I've decided to stop this for inside the function and not execute it at all. Then I've printed out the length in both calls of compareRows and they're being printed correctly. Then instead of doing that for, I've decided to put a for that does nothing, just iterate N times. And for my surprise, the length get corrupted in the second call when N >= 985. Even more weird, if I print the length before using compareRows in the second time then it'll show the correct length... it just happens when it goes inside the function. I've also attached the modified version of this test that I was using to debug. My output for this debug: typeof(ctx) = object # comparing data... size1=240000 # comparing data... size2=240000 comparing data... size0=240000 comparing data... size1=240000 ## typeof(ctx) = object ## comparing data... size1=240000 ## comparing data... size2=240000 # comparing data... size1=623014176 # comparing data... size2=623014176 comparing data... size0=623014176 comparing data... size1=623014176

This appears to time out on all platforms. The test should be skipped/disabled.

(In reply to comment #3) > This appears to time out on all platforms. The test should be skipped/disabled. I've imagined that... the issue would just disappear depending on what context (inside a function or not) the data comparison is done, this bothers me. It could be slightly modified to pass... instead of creating a function to avoid code repetition and compare the image data, we could avoid creating that function. This approach would probably solve it but would also hide this issue that is potentially something big... I'm sorry I couldn't investigate more about it.

(In reply to comment #0) > $ Tools/Scripts/old-run-webkit-tests fast/canvas/canvas-lineWidth.html --iterations 100 --singly For the record: this doesn't timeout for me neither in Qt on Mac Lion (64 bits) and Apple port on Mac Lion.

(In reply to comment #5) > (In reply to comment #0) > > $ Tools/Scripts/old-run-webkit-tests fast/canvas/canvas-lineWidth.html --iterations 100 --singly > > For the record: this doesn't timeout for me neither in Qt on Mac Lion (64 bits) and Apple port on Mac Lion. It passes on Qt Linux now ...

It looks like someone has unskipped it already, then I will close this bug.