945811 - Assertion failure: !analysisContext->isExceptionPending(), at jit/IonBuilder.cpp:3868

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Jan de Mooij [:jandem]]

Attachment: Patch

decoder reported this on IRC. The problem was that the call to getOrCreateScript in IonBuilder was not propagating an error.

This patch refactors the inlining code a bit so that this is handled correctly.

This may fix intermittent orange bug 939714 and/or bug 923614.

Comment 1

[Jan de Mooij [:jandem]]

Attachment: Patch

Forgot to add the testcase.

Comment 2

[Hannes Verschore [:h4writer]]

Comment on attachment 8341804 [details] [diff] [review]
Patch

Review of attachment 8341804 [details] [diff] [review]:
-----------------------------------------------------------------

Nice. We indeed need to have some error propagation there.
Though I have a few comments:

1) Why don't we reuse InliningStatus, instead of adding InliningDecision. I don't see any reason not to. They both have the same 3 values (error, inlined, notInlined). No need for this extra enum. Though I wouldn't be mad if we decided to rename InliningStatus to InliningDecision ;)

2) IonBuilder::DontInline should be IonBuilder::dontInline(targetScript, const char *reason). Lowercase.

Looks much cleaner with the dontInline spewing ;)

Ps: did you have to do this before optimizationLevels landed :P

Comment 3

[Jan de Mooij [:jandem]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/48b19f9cc967

(In reply to Hannes Verschore [:h4writer] from comment #2)
> 1) Why don't we reuse InliningStatus, instead of adding InliningDecision.

Discussed this on IRC - it felt wrong to use "InliningStatus_Inlined" when we are not sure we are going to inline..

> 
> 2) IonBuilder::DontInline should be IonBuilder::dontInline(targetScript,
> const char *reason). Lowercase.

It's a static method so it has to be uppercase - see also MBasicBlock::New, IonBuilder::TableSwitch, etc.

> Ps: did you have to do this before optimizationLevels landed :P

Sorry! For some reason we're touching the same code lately :(

Comment 4

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/48b19f9cc967