Set LD=ld.bfd on Alpine linux during bindist configure

[hasufell]

Fixes #967

[runeksvendsen]

How much work would it be to test in CI that this fixes the segfault issue?

[runeksvendsen]

This also fixes #962, right?

[hasufell]

This also fixes #962, right?

No, it unfortunately doesn't. I'm suspecting an actual bug in 9.4.8 series.

[hasufell]

How much work would it be to test in CI that this fixes the segfault issue?

I don't feel that's gonna add much.

We have evidence here:

• https://gitlab.haskell.org/ghc/ghc/-/issues/17508#note_313450
• https://sourceware.org/bugzilla/show_bug.cgi?id=23856

The minimal reproducer (C code) is shown here (and still works): https://gitlab.haskell.org/ghc/ghc/-/issues/17508#note_372662

There's no minimal reproducer for Haskell.

My past days as Gentoo distributor also remind me that ld.gold was always more problematic, so I don't think it's a good default to begin with.

[hasufell]

There's no minimal reproducer for Haskell.

Ah, silly me... I didn't edit the settings file correctly.

Obviously, this works:

$ cat Foo.hs
main = putStrLn "hello world"
ghc Foo.hs
./Foo
Segmentation fault (core dumped)

Then make sure your ghc settings file has these lines:

,("C compiler link flags", "-fuse-ld=gold")
,("ld command", "ld.gold")
,("Merge objects command", "ld.gold")

[runeksvendsen]

The way I read the code, we previously ran the configure script with an empty environment — ie. it did not inherit anything from the environment of GHCup. Is this correctly understood?

If this is the case then I'm not convinced that we should pass all of GHCup's environment plus LD=ld.bfd to the configure script in the positive case. In that case why not use an environment with only LD=ld.bfd?

[hasufell]

we previously ran the configure script with an empty environment — ie. it did not inherit anything from the environment of GHCup. Is this correctly understood?

No, that's not correct. That parameter overwrites the current environment for the subprocess. LD=ld.bfd ghcup install ghc ... works perfectly fine.

[runeksvendsen]

we previously ran the configure script with an empty environment — ie. it did not inherit anything from the environment of GHCup. Is this correctly understood?

No, that's not correct. That parameter overwrites the current environment for the subprocess. LD=ld.bfd ghcup install ghc ... works perfectly fine.

So passing Just [("LD", "ld.bfd")] as the last argument to execLogged will execute the process in the same environment as that of GHCup but with the LD env var overwritten?

If that's the case then why do we need

(("LD", "ld.bfd") : cEnv)

instead of just

[("LD", "ld.bfd")]

?

[hasufell]

So passing Just [("LD", "ld.bfd")] as the last argument to execLogged will execute the process in the same environment as that of GHCup but with the LD env var overwritten?

No, you pass the entire environment.

[runeksvendsen]

Okay I understand now. I thought passing Nothing as the environment means the executed process would have an empty environment. But Nothing means "inherit environment of the parent process" whereas Just [] means "use an empty environment".

[runeksvendsen]

The code looks good to me. But I don't have a way to test it, so I will leave that up to you.