More windows woe

[Ben Gamari]

Simon Peyton Jones via ghc-devs <ghc-devs at haskell.org> writes:

> I see that anything involving ghci fails:
>
> /c/code/HEAD/inplace/bin/ghc-stage2 --interactive
>
> GHCi, version 8.1.20161209: http://www.haskell.org/ghc/  :? for help
>
> ghc-stage2.exe: unable to load package `base-4.9.0.0'
>
> ghc-stage2.exe: C:\code\HEAD\inplace\mingw\x86_64-w64-mingw32\lib\libmingwex.a: unknown symbol `_lock_file'
>
Yes, Tamar and I were working on tracking this down over the last few
days. The patch (which I will merge after a running validation finishes)
is D2817.

In short, the problem is that we recently upgraded the Windows toolchain.
For better or worse, the new mingw-w64 toolchain now has an atomic
printf implementation, which requires the use of the _lock_file function
provided by Microsoft's C runtime. However, the _lock_file symbol is
only exported by certain variants of msvcrt (e.g. msvcrt90.dll), but not
the distribution which mingw-w64 uses (apparently due to license
considerations [1], although the exact reason isn't clear).

To hack around this, mingw-w64 ships a static library, msvcrt.a, which
wraps msvcrt.dll and provides hand-rolled implementations of some needed
symbols, including _lock_file. However, this means that the static
library msvcrt.a, and the dynamic library msvcrt.dll don't export the
same set of symbols, which causes GHCi to blow up if dynamically linked.
Consequently we need to 

All of this coupled with another recent but quite unrelated cleanup
(D2579) breaking the Windows build when bootstrapped with GHC 7.10, the
recent testsuite debacle, as well as a number of other Windows quirks
I've discovered in the past few weeks, meant that figuring all of this
out took quite some time (which is why the Windows builder *still* isn't
quite up). On the bright side, one happy side-effort of this is that it
prompted me to write down some notes on the interactions between the
many components of our Windows toolchain [2].

Anyways, we are getting quite close. I expect we'll finally have the
Windows builder up by next week. Hopefully from that point forth it will
be considerably harder to break the Windows build.

Cheers,

- Ben


[1] https://sourceforge.net/p/mingw-w64/discussion/723797/thread/55520785/
[2] https://ghc.haskell.org/trac/ghc/wiki/SurvivingWIndows
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 454 bytes
Desc: not available
URL: <http://mail.haskell.org/pipermail/ghc-devs/attachments/20161209/8457b7a0/attachment.sig>