REGRESSION (r72351): fast/profiler/throw-exception-from-eval.html fails

• Component: New Bugs
• Status: RESOLVED INVALID

[Csaba Osztrogonác]

(not so) pretty diff:
http://build.webkit.org/results/Qt%20Linux%20Release/r72386%20%2823991%29/fast/profiler/throw-exception-from-eval-pretty-diff.html

diff apart from whitespace differences:
$ diff -Nurw throw-exception-from-eval-expected.txt  throw-exception-from-eval-actual.txt
--- throw-exception-from-eval-expected.txt      2010-11-19 06:02:08.602130028 -0800
+++ throw-exception-from-eval-actual.txt        2010-11-19 06:02:08.602130028 -0800
@@ -5,7 +5,6 @@

 Profile title: Throw within an eval.
 Thread_1 (no file) (line 0)
-   (program) throw-exception-from-eval.html (line 19)
       (program) throw-exception-from-eval.html (line 10)
          (program) throw-exception-from-eval.html (line 4)
          (program) throw-exception-from-eval.html (line 19)

[Csaba Osztrogonác]

I added the failing test to the Skipped list: http://trac.webkit.org/changeset/72397

[Andy Estes]

These tests also fail on the Leopard bots.

[Andy Estes]

Skipped on Leopard bots in http://trac.webkit.org/changeset/72406.

[Alexey Proskuryakov]

See also: bug 49809.

[Gavin Barraclough]

I'm not sure what these tests are supposed to output, but my guess here is that this is a progression, and that the new Qt/Leopard results are correct.

Looking at the results, they are showing 'line 19' at the top and bottom of the call stack.  Looking at the .html source, this corresponds to a script tag containing an eval.  It is unlikely that there is a reason for this to be at both ends of the chain, so I think this indicates an error previously in failing to remove layers from the call stack upon exception unwind, where they should have been being removed.

I'll have to investigate more to confirm this.

[Alejandro G. Castro]

It also fails in gtk+, I'm skipping it until we find the issue.

[Jessie Berlin]

(In reply to comment #6)
> It also fails in gtk+, I'm skipping it until we find the issue.

It also fails on Win-XP. I will check in the failing results soon.

[Gavin Barraclough]

Hi Alejandro,

This problem relates to:
    https://bugs.webkit.org/show_bug.cgi?id=26276
If GTK is testing on a platform that is not yet correctly measuring its stack size, then fixing this will fix the problem.

However we may have to expect different results here.  Is it also possible that GTK is testing on a platform where the stack extent is being measured correctly, but the machine stack is smaller.  If this is the case, the test will not be able to recurse as many times, and will be expected to produce different results.

cheers,
G.

[Adam Roben (:aroben)]

(In reply to comment #7)
> (In reply to comment #6)
> > It also fails in gtk+, I'm skipping it until we find the issue.
> 
> It also fails on Win-XP. I will check in the failing results soon.

Jessie checked in failing results in r74560. The Windows XP bots are now not matching those failing results anymore; they're matching the cross-platform results: <http://build.webkit.org/results/Windows%20XP%20Debug%20(Tests)/r75711%20(24147)/fast/profiler/throw-exception-from-eval-pretty-diff.html>.

It looks like the test produces the Windows-XP-specific results when run on its own, and produces the cross-platform results when run after other tests in fast/profiler. I haven't been able to determine a single other test that makes this one start producing cross-platform results.

[Adam Roben (:aroben)]

(In reply to comment #9)
> It looks like the test produces the Windows-XP-specific results when run on its own, and produces the cross-platform results when run after other tests in fast/profiler. I haven't been able to determine a single other test that makes this one start producing cross-platform results.

Even worse, it doesn't seem to be deterministic. This command produced different results when run two times in a row:

run-webkit-tests fast/profiler/anonymous-event-handler.html fast/profiler/throw-exception-from-eval.html

[Adam Roben (:aroben)]

(In reply to comment #10)
> Even worse, it doesn't seem to be deterministic. This command produced different results when run two times in a row:
> 
> run-webkit-tests fast/profiler/anonymous-event-handler.html fast/profiler/throw-exception-from-eval.html

Given the non-deterministic failure, I think it makes sense to disable this test on all platforms.

[Adam Roben (:aroben)]

(In reply to comment #11)
> Given the non-deterministic failure, I think it makes sense to disable this test on all platforms.

Disabled the test in r75723.

[Adam Roben (:aroben)]

<rdar://problem/8860201>

[WebKit Review Bot]

http://trac.webkit.org/changeset/75723 might have broken SnowLeopard Intel Release (Tests)
The following tests are not passing:
fast/css/pseudo-in-range-invalid-value.html
fast/css/pseudo-in-range.html
fast/forms/form-collection-elements.html
fast/forms/range-keyoperation.html

[Joseph Pecoraro]

Legacy Profiler has been removed. This test no longer exists.