584354 - Assertion failure: JS_HAS_OPTION(acx, JSOPTION_UNROOTED_GLOBAL) whilst running Thunderbird message-header tests in debug mode

Status: RESOLVED DUPLICATE of bug 584495

(Core :: XPCOM, defect)

Description

[Mark Banner (:standard8)]

Attachment: backtrace

I'm trying to run the Thunderbird Mozmill-based message-header tests in debug mode. I'm getting:

Assertion failure: JS_HAS_OPTION(acx, JSOPTION_UNROOTED_GLOBAL), at /Users/moztest/comm/main/src/mozilla/js/src/xpconnect/src/xpcjsruntime.cpp:383

it happens on every run, but in different places, which seems to indicate it is cycle collection related.

I'm flagging this as a possible regression from bug 549806 as that is the bit that touched this code last.

Also attaching a backtrace at the point the assertion occurs.

Comment 1

[Mark Banner (:standard8)]

Note: if I make the assertion into an if statement, then it all runs fine - though obviously there could be some hidden issue.

Comment 2

[Mark Banner (:standard8)]

Attachment: Workaround - change assertion back to if.

I have no idea if this is right or not, but it works around the issue - it puts back the if statement that was there before the assertion.

As other Thunderbird developers are seeing this issue I'm putting this up here as a starter for ten.

Comment 3

[Brendan Eich [:brendan]]

This needs to be fixed. Assertions are fatal, or should be, and they're treated as critical bugs anwyay (in SpiderMonkey). No fair leaving them in Gecko code to bite others.

/be

Comment 4

[Andreas Gal :gal]

Mark do you have a way to reproduce this?

Comment 5

[Mark Banner (:standard8)]

Andreas, unfortunately the only way I've found so far is:

1) Build Thunderbird in debug mode (https://developer.mozilla.org/En/Simple_Thunderbird_build) with enable-tests

2) Have Mozmill installed: https://developer.mozilla.org/en/Thunderbird/Thunderbird_MozMill_Testing

3) Run "make SOLO_TEST=message-header mozmill-one" from objdir.

Wait for it to crash (although I've found I can hook in the debugger before the cycle collection happens).

I can try and find a simpler way next week, but I'm not confident of finding one.

Comment 6

[Mark Banner (:standard8)]

I've run this a few times now with the latest tracemonkey merge and I can't get it to fail. Therefore I think this is a duplicate of bug 584495.