Bug 25292 – Apparent memory corruption running fast/js/math-transforms.html

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Gustavo Noronha (kov)]

Trying to debug two failing tests on the GTK+ port I found out that they are failing because the frame name goes from empty to a very strange value: "someFloatString". This is caused by running the math-transforms test. I added small print statements to the beginning and the end of the FrameLoader::executeScript method to print the frame name, to make sure and this is the output I get when I navigate to that test using the launcher:

name (executescript): 
name (executescript2): 
name (executescript): 
name (executescript2): someFloatString
name (executescript): someFloatString
name (executescript2): someFloatString

This happens both in GTK+ and Qt. Haven't been able to try other platforms yet. I have used the print statements because I wasn't able to figure out a better way of tracking/nailing this problem; valgrind didn't show any suspicious writes, and I was unable to figure out a watch statement that would catch this in gdb. I am on Linux x86, and I have JIT enabled.

[Gustavo Noronha (kov)]

Created attachment 29834 [details]
gdb output of investigation

I am still investigating this problem. Still not sure I'm on the right track, but I thought I'd post this output here, for it may be useful for someone familiar with the code.

[Jan Alonzo]

Created attachment 31094 [details]
clear the frame name before running each test

someFloatString is apparently the frame name, so we need to make sure we clear the frame name before we run each test. This is also what mac does by the way.

[Gustavo Noronha (kov)]

Comment on attachment 31094 [details]
clear the frame name before running each test

Since Mac is also doing it, sounds fine to me. But I wonder if we should not be clearing this as part of the normal load process?

[Brent Fulgham]

Landed in @r44570.

[Brent Fulgham]

Landed in @r44570.

[Gustavo Noronha (kov)]

The DRT part was missing in the first commit. I just landed it in r44594.