Duplicate symbol errors when running GHCi/TH tests (libffi pains)

[hasufell]

• Branch: https://github.com/stable-haskell/ghc/tree/release-ci
• PR: Release CI #75

Problem

GHC runtime linker: fatal error: I found a duplicate definition for symbol
   ffi_prep_cif
whilst processing object file
   /usr/home/github/.cache/act/aa026268a9c70a03/hostexecutor/_build/bindist/lib/x86_64-portbld-freebsd/libffi-3.5.2/HSlibffi-3.5.2.o
The symbol was previously defined in
   (GHCi built-in symbols)
This could be caused by:
   * Loading two different object files which export the same symbol
   * Specifying the same object file twice on the GHCi command line
   * An incorrect `package.conf' entry, causing some object to be
     loaded twice.
<no location info>: error:
    loadObj "/usr/home/github/.cache/act/aa026268a9c70a03/hostexecutor/_build/bindist/lib/x86_64-portbld-freebsd/libffi-3.5.2/HSlibffi-3.5.2.o": failed

Full log: https://github.com/stable-haskell/ghc/actions/runs/18005814665/job/51230757936#step:10:2023

Explanation

This bug was originally introduced by using a libffi Haskell package, which was added here: 898758a

The reason for this was that we don't ship a bundled libffi anymore like the vanilla bindists. This causes problems when linking on mac and FreeBSD: now the user has to manually tell cabal/ghc where libffi lies.

The duplicate symbol error was fixed by this hack in rts/Linker.c: https://github.com/stable-haskell/ghc/compare/8f643e5d60376266cbaa27c7fba23eca1f2c2522%5E..ce3e1603ccb748062400c21d33780e6e7142d502#diff-6d91534f6d8d658a4dbb8ab58c56b755c2f115f7abb7952c4f85377e6aabc966

Then later we tried to fix it more cleanly with @angerman: 4ba4451

But it turned out this doesn't work. CI is still failing.

[hasufell]

This seems to have fixed it for me: ac90c87