960544 - GenerationalGC: New failures in cppunittests that link to JS

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jon Coppeard (:jonco)]

The following cppunittest failures have started appearing in generational GC browser builds:

Assertion failure: initialized(), at ../../../../dist/include/mozilla/ThreadLocal.h:123
cppunittests TEST-UNEXPECTED-FAIL | test_IHistory | test failed with return code -11
Assertion failure: initialized(), at ../../../../dist/include/mozilla/ThreadLocal.h:123
cppunittests TEST-UNEXPECTED-FAIL | sdp_unittests | test failed with return code -11
Assertion failure: initialized(), at ../../../../dist/include/mozilla/ThreadLocal.h:123
cppunittests TEST-UNEXPECTED-FAIL | mediapipeline_unittest | test failed with return code -11
Return code: 1

I suspect this is fallout from bug 951743 as these tests all end up linking to the JS library.

Comment 1

[Jon Coppeard (:jonco)]

Attachment: bug960544-linking-unittests

Right, so this is the problem mentioned in bug 951743 comment 6.

Now that JS_FRIEND_API always exports, we can use that for the relocation functions that these test programs need and stop linking them twice against the JS library.

Comment 2

[Mike Hommey [:glandium]]

Comment on attachment 8361167 [details] [diff] [review]
bug960544-linking-unittests

Review of attachment 8361167 [details] [diff] [review]:
-----------------------------------------------------------------

Thank you

Comment 3

[Jon Coppeard (:jonco)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/942064e2a64e

Comment 4

[Jon Coppeard (:jonco)]

It looks like this change breaks GGC builds on windows, which can't find the relocation functions:

test_IHistory.obj : error LNK2019: unresolved external symbol "__declspec(dllimport) void __cdecl JS::HeapCellRelocate(struct js::gc::Cell * *)" (__imp_?HeapCellRelocate@JS@@YAXPAPAUCell@gc@js@@@Z) referenced in function "public: static void __cdecl js::GCMethods<class JSObject *>::relocate(class JSObject * *)" (?relocate@?$GCMethods@PAVJSObject@@@js@@SAXPAPAVJSObject@@@Z)

test_IHistory.obj : error LNK2019: unresolved external symbol "__declspec(dllimport) void __cdecl JS::HeapCellPostBarrier(struct js::gc::Cell * *)" (__imp_?HeapCellPostBarrier@JS@@YAXPAPAUCell@gc@js@@@Z) referenced in function "public: static void __cdecl js::GCMethods<class JSObject *>::postBarrier(class JSObject * *)" (?postBarrier@?$GCMethods@PAVJSObject@@@js@@SAXPAPAVJSObject@@@Z)

test_IHistory.exe : fatal error LNK1120: 2 unresolved externals

Comment 5

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/942064e2a64e

Comment 6

[Jon Coppeard (:jonco)]

Ok, so I don't really understand what's going on here.  My guess would be that these functions are not being exported on Windows for some reason.

I seem to recall that linking is done differently on Windows though, is that true?  If you've got time to explain to me how all this works that would be much appreciated.

Comment 7

[Mike Hommey [:glandium]]

I can't really tell without looking, and i can't build m-c with generational gc enabled, it fails to build well before your failure.

Comment 8

[Mike Hommey [:glandium]]

My tree might have been too old (less than 2 days old)... it's building now.

Comment 9

[Mike Hommey [:glandium]]

Ah I see what's going on. The test is using symbols from the js library, but on windows, it's *not* in libxul. Patch incoming.

Comment 10

[Mike Hommey [:glandium]]

Attachment: Link TestIHistory against mozjs when it is a shared library

Comment 11

[Mike Hommey [:glandium]]

Attachment: Link some cppunittests against mozjs when it is a shared library

With a more appropriate message.

Comment 12

[Jon Coppeard (:jonco)]

(In reply to Mike Hommey [:glandium] from comment #9)
Great, thank you for looking into this!

Comment 13

[Mike Hommey [:glandium]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/dd6721572000

Comment 14

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/dd6721572000