732308 - xpcshell/head.js leaks component manager in some cases

Status: RESOLVED FIXED

(Testing :: XPCShell Harness, defect)

Description

[Luke Wagner [:luke]]

Attachment: fix

A side-effect of bug 730497 is to make do_get_profile (testing/xpcshell/head.js:907) a "heavyweight" function.  This causes the component manager's refcount to not go all the way down to 0 in all the services/sync xpcshell tests.  However, this is nothing special about bug 730497, to see this on trunk, just put an "eval('')" at the head of do_get_profile and run 'make xpcshell-tests'.  (You could also just run any individual sync test, say, test_Observers.js, but bug 732255 won't pass the -d flag to xpcshell which is what turns on the assert-crash.)

The actual problem seems to be some unnatural interaction between the lifetime of the component manager and xpcshell.  Nulling out the unnecessary vars in the newly-entrained scope fixes the problem.

Comment 1

[Luke Wagner [:luke]]

Comment on attachment 602246 [details] [diff] [review]
fix

Changing r?jmaher at ted's suggestions.

Comment 2

[Joel Maher ( :jmaher ) (UTC -8)]

Comment on attachment 602246 [details] [diff] [review]
fix

Review of attachment 602246 [details] [diff] [review]:
-----------------------------------------------------------------

these changes look good

Comment 3

[Luke Wagner [:luke]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/dfa8feda90d6

Comment 4

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

https://hg.mozilla.org/mozilla-central/rev/dfa8feda90d6

Comment 5

[Luke Wagner [:luke]]

For posterity, from irc, running the command: 

  python -u ../config/pythonpath.py  -I../build  ../testing/xpcshell/runxpcshelltests.py  --build-info=./mozinfo.json  dist/bin/xpcshell  _tests/xpcshell/js/xpconnect/tests/unit --test-path=test_bug604362.js

does not pass -d (the test doesn't want it), but 'make xpcshell-tests' passes -d for this test.

Comment 6

[Luke Wagner [:luke]]

Oops, wrong bug, ignore that.