755163 - Some jit tests fail with compiler optimizations disabled

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Comment 1

[:Benjamin Peterson]

Specifically, configure with --disable-optimize and run jit tests to get

FAILURES:
    -m /home/benjamin/dev/repos/y/js/src/jit-test/tests/basic/bug529130.js
    -m -n /home/benjamin/dev/repos/y/js/src/jit-test/tests/basic/bug529130.js
TIMEOUTS:

Comment 2

[David Mandelin [:dmandelin]]

It passes for me on 32-bit Windows and MacOS 10.6. What version of GCC are you using?

Comment 3

[:Benjamin Peterson]

gcc 4.5.3 on Linux.

Comment 4

[:Benjamin Peterson]

The output looks like

|    0.0s1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
1,2,3,4,5,6,7,8,9
/home/benjamin/dev/repos/y/js/src/jit-test/tests/basic/bug529130.js:6: InternalError: too much recursion
Exit code: 3

Comment 5

[Jeff Walden [:Waldo]]

That's probably because the test recurs a lot, and usable recursion depth depends on the stack space consumed by the compiler.  I consider this test flaky and ignore failures from it, which quite arguably I shouldn't do.  Maybe we should actually fix the test to not recur so close to the overall limit that compiler-dependent stack sizes makes it fail?

Comment 6

[:Benjamin Peterson]

Is this test even very useful? The code for the bug that it fixes is gone with tracing.

Comment 7

[David Mandelin [:dmandelin]]

(In reply to Benjamin Peterson from comment #6)
> Is this test even very useful? The code for the bug that it fixes is gone
> with tracing.

Kill it with fire!

Comment 8

[:Benjamin Peterson]

Attachment: kills the offending test

Comment 9

[:Benjamin Peterson]

Attachment: kills offending test

Comment 10

[Paul Wright]

Is this the only test which exercises code which is no longer in use?  Should there be some audit activities to find out?

Comment 11

[:Benjamin Peterson]

(In reply to Paul Wright from comment #10)
> Is this the only test which exercises code which is no longer in use? 
> Should there be some audit activities to find out?

I'm not sure what you mean. The code which this test tested is gone.

Comment 12

[Jeff Walden [:Waldo]]

w00t!  Always did find that test annoying.

Comment 13

[Jeff Walden [:Waldo]]

...ahem:

https://hg.mozilla.org/integration/mozilla-inbound/rev/e1406f8b5d54

Comment 14

[Paul Wright]

(In reply to Benjamin Peterson from comment #11)
> (In reply to Paul Wright from comment #10)
> > Is this the only test which exercises code which is no longer in use? 
> > Should there be some audit activities to find out?
> 
> I'm not sure what you mean. The code which this test tested is gone.

Exactly.  This test tested code which is now gone.  Are there any other tests which also test code which no longer exists?  Is there a way to audit the test suites to find out / clean them up?

Comment 15

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/e1406f8b5d54

Leaving open as it's not clear whether more work is planned for this bug or not. Please resolve it if there isn't.

Comment 16

[:Benjamin Peterson]

No tests failure anymore, so this bug specifically is done.