1141660 - Total build bustage when Gecko 39 merges to Aurora

Status: VERIFIED FIXED

(Core :: Memory Allocator, defect)

Description

[Ryan VanderMeulen [:RyanVM]]

[Tracking Requested - why for this release]: Complete build bustage when Gecko 39 merges to Aurora.

bz was hitting this locally as well when trying to reproduce test failures I was hitting on an uplift simulation Try push from yesterday. Something keying off NIGHTLY_BUILD that breaks when it isn't defined? This blocks me from running uplift simulation pushes, so please take a look ASAP.

https://treeherder.mozilla.org/logviewer.html#?job_id=5530249&repo=try

In file included from /builds/slave/try-and-api-9-0000000000000000/build/obj-firefox/memory/mozalloc/Unified_cpp_memory_mozalloc0.cpp:2:0:
/builds/slave/try-and-api-9-0000000000000000/build/memory/mozalloc/mozalloc.cpp:29:26: fatal error: malloc_decls.h: No such file or directory
compilation terminated.
make[5]: *** [Unified_cpp_memory_mozalloc0.o] Error 1
make[4]: *** [memory/mozalloc/target] Error 2
make[4]: *** Waiting for unfinished jobs....

Comment 1

[Boris Zbarsky [:bzbarsky]]

So the reason this works on nightly is that memory/build/moz.build exports malloc_decls.h if MOZ_REPLACE_MALLOC and we always define MOZ_REPLACE_MALLOC if NIGHTLY_BUILD in configure.in.

So the difference is that in a non-nightly malloc_decls.h is not exported.  

mozalloc.cpp is in memory/mozalloc, not memory/build.  And it didn't use to include malloc_decls.h until bug 868814.

Comment 2

[Mike Hommey [:glandium]]

Attachment: Fix bustage from bug 868814 with --disable-replace-malloc

Comment 3

[Nicholas Nethercote [inactive]]

Comment on attachment 8575654 [details] [diff] [review]
Fix bustage from bug 868814 with --disable-replace-malloc

Review of attachment 8575654 [details] [diff] [review]:
-----------------------------------------------------------------

rs=me

Comment 4

[Mike Hommey [:glandium]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/1f34ebc9eb00

Comment 5

[Mike Hommey [:glandium]]

That's not the end of it. The following happens on windows, now:

mozmemory_wrap.obj : error LNK2019: unresolved external symbol _je_memalign referenced in function __aligned_malloc
mozglue.dll : fatal error LNK1120: 1 unresolved externals
mozmake.exe[5]: *** [mozglue.dll] Error 1120
mozmake.exe[4]: *** [mozglue/build/target] Error 2
mozmake.exe[4]: *** Waiting for unfinished jobs....

Comment 6

[Mike Hommey [:glandium]]

That error in comment 5 comes from bug 762449

Comment 7

[Mike Hommey [:glandium]]

Attachment: Always export memalign from jemalloc, even on Windows

Comment 8

[Nicholas Nethercote [inactive]]

Comment on attachment 8575739 [details] [diff] [review]
Always export memalign from jemalloc, even on Windows

Review of attachment 8575739 [details] [diff] [review]:
-----------------------------------------------------------------

rs=me

Comment 9

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/1f34ebc9eb00

Comment 10

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/80555e0558a3