1154596 - [gcc 5.0] gmake[6]: *** [test-ctors.so] Error 1 when building with -flto

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Mike Hommey [:glandium]]

21:00:41 INFO - # Fail if the library doesn't have INIT .dynamic info
21:00:41 INFO - readelf -d test-ctors.so | grep '(INIT)'
21:00:41 INFO - gmake[6]: *** [test-ctors.so] Error 1

Comment 1

[Mike Hommey [:glandium]]

Attachment: Build elfhack tests with -fno-lto

Comment 2

[Nathan Froyd [:froydnj]]

Comment on attachment 8592626 [details] [diff] [review]
Build elfhack tests with -fno-lto

Review of attachment 8592626 [details] [diff] [review]:
-----------------------------------------------------------------

I find it a little odd that the checkin comment justifies the solution as "more correct" when it's fixing "assumptions" made by test-ctors, but whatever.

::: build/unix/elfhack/moz.build
@@ +15,5 @@
>      ]
>  
> +    if '-flto' in CONFIG['OS_CXXFLAGS']:
> +        SOURCES['test-array.c'].flags += ['-fno-lto']
> +        SOURCES['test-ctors.c'].flags += ['-fno-lto']

This construct reads a little funny: if we're LTO'ing C++, don't LTO these C files?  Can we check for -flto in OS_CFLAGS instead, to make things consistent?  Or maybe OS_LDFLAGS would be even better?

Comment 3

[Trevor Saunders (:tbsaunde)]

> I find it a little odd that the checkin comment justifies the solution as
> "more correct" when it's fixing "assumptions" made by test-ctors, but
> whatever.

Well, correct is tricky to define right?  Since the point of elf hack is to fix up binaries the toolchain layed out stupidly in some sense it makes sense to test it on binaries that are layed out that way.  On the other hand we want to test elf hack is doing something useful, so it makes sense to test it on code layed out similarly to how real binaries will be layed out.  If the layout has changed it seems like it might be worth investigating if elf hack still helps, and if not if that's because its unnecessary or if its just not working properly on the new layout.

Comment 4

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/8e0a4a2295f8

Comment 5

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/8e0a4a2295f8