Re: Crowdfunding a small but important powerpc bugfix
Hello!
On Wed, 2023-10-11 at 15:38 +0200, John Paul Adrian Glaubitz wrote:
> There is a regression in the GHC Haskell Compiler [1] that I have been
> investigating for several days. While I have been able to identify the
> commit that introduced the regression, I have not been able to fix
> that issue since I don't have the necessary Haskell knowledge to be able
> to work on GHC.
>
> Since the broken GHC compiler currently blocks a lot of packages on 32-bit
> PowerPC and the upstream developers don't seem to be too motivated at the
> moment to fix the bug, I was wondering whether people here would be willing
> to throw in some money to create a bounty on that particular bug to pay
> a Haskell developer to fix it.
We're starting with good news into the new year: I have been able to work around
this bug and successfully built an unregisterised version of GHC 9.4.6 for powerpc.
I have uploded ghc_9.4.6-1+powerpc this morning to "unreleased" and the buildds are
now building all Haskell packages on powerpc that were out of date [1].
I was able to build an unregisterised compiler by setting the GCC optimization level
to "O3". Building an unregisterised compiler on powerpc previously failed due to the
generated C code being too large for the powerpc target [2]. The suggested CFLAGS by
Segher in the comments of the GCC bug report did the trick.
If we're lucky, the original regression of the 32-bit PowerPC native codegen backend
might already be fixed in newer versions of the GHC compiler [3], so that we can switch
back from an unregisterised build to the faster, native build of GHC. We'll see.
Adrian
> [1] https://buildd.debian.org/status/architecture.php?a=powerpc&suite=sid
> [2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108208
> [3] https://gitlab.haskell.org/ghc/ghc/-/issues/23969
--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
Reply to: