Firefox segfaults on launch

[dylanaraps]

The last good commit (which produces a working Firefox) is b99626f. All subsequent commits (that I have tried) produce a Firefox binary which segfaults on startup. It seems as though 475a250 is the cause though I have no proof to back this up. I will be going through commit by commit to see exactly where the issue is introduced. This will take me some time as it's 3+ hours per build.

[dylanaraps]

Let me know if there is any information I can provide also.

[dylanaraps]

OK. 0.9.4 produces a segfaulting binary.

One of these two commits introduces the issue(?):

• 06b11da
• 475a250

[rui314]

Thank you for reporting! It is very likely to be caused by 475a250. How did you build Firefox? Are you building a release binary or a debug?

[dylanaraps]

I built a release binary.

Build instructions: https://github.com/kisslinux/repo/blob/master/testing/firefox/build
CFLAGS/CXXFLAGS: -O3 -march=native -pipe

[rui314]

Thank you for sharing the build instruction. I succeeded to reproduce the issue. Investigating...

This is a major regression, so I'll probably create a new release after fixing this bug.

[rui314]

This comment states that Firefox uses a linker "loophole" that it will promote weak undef symbols to dynamic symbols even if -z defs is given when building a .so file. This loophole is exactly what I wanted to plug as I thought it was more like a bug than a feature. But it looks like I can't do that. I have to make mold compatible with mozilla.

https://hg.mozilla.org/mozilla-central/file/tip/mfbt/Types.h#l91

[rui314]

I believe I managed to fix this. Could you verify if you have time?

[dylanaraps]

Yes. I will verify now and let you know in 2-3 hours (edit: Will be tomorrow).

[dylanaraps]

Works, thanks. :)