[GHC DevOps Group] CircleCI status

Phyx lonetiger at gmail.com
Sat Dec 22 18:05:15 UTC 2018

On Sat, Dec 22, 2018 at 5:40 PM Ben Gamari <ben at well-typed.com> wrote:

> Ccing our Windows czar, Tamar.
> "Greg Steuck (Sh-toy-k)" <gnezdo at google.com> writes:
> > FWIW, lld seems pretty well supported on Windows
> > <https://lld.llvm.org/windows_support.html>. We have some evidence that
> lld
> > is faster than gold and ld.bfd on Linux.
> >
> Indeed that is the option that I am investigating. I have opened #16084
> [1] and an msys2 [2] ticket to track this. Unfortunately, GCC shipped
> with msys2 doesn't currently support -fuse-ld=lld, so we are stuck with
> ld.bfd until GCC 9 is released. However, even then it sounds like
> lld <8 and mingw aren't known to interact well.

I wasn't even aware they added a mingw front-end back into LLD, this does
make it a viable option.
That said I also have a somewhat optimized binutils port here

The patch for fuse-ld=lld support is a trivial backport. I can easily build
a GCC 8 release with it included and
a tip of trunk LLD 8 if you'd like to try them out. I wouldn't recommend
GCC 9 atm as we're in stage 3 and things
are routinely broken..

> All-in-all, moving to lld sounds like a good long-term direction but
> won't be viable for at least a few months.

Yes adding it as an extra option and maybe default is good, especially as
it has PDB support, making life on Windows
much much easier, but we should keep bfd support as well as bfd tends to
have better obscure feature support, for those that need it.


> Cheers,
> - Ben
> [1] https://ghc.haskell.org/trac/ghc/ticket/16084
> [2] https://github.com/Alexpux/MSYS2-packages/issues/1531
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-devops-group/attachments/20181222/e8924a6b/attachment.html>

More information about the Ghc-devops-group mailing list