Re: MPS: Win64 testers?

[Pip Cet]

"Eli Zaretskii" <eliz@gnu.org> writes:
>> From: Quang Kien Nguyen <kien.n.quang@gmail.com>
>> Date: Tue, 13 Aug 2024 05:12:19 -0700
>> Cc: emacs-devel@gnu.org
>>
>> By the way, @Pip Cet, have you been able to build 64bit Emacs with MPS?

(I assume you mean for Microsoft Windows).

Yes, I have.  It's even started on a machine running Microsoft Windows,
though I'm usually limited to testing on wine, and even that binary was
built using wine.

>> I've tried to build them but Emacs crashed as soon as temacs was
>> trying to bootstrap.

That's probably the setjmp thing, worked around in my tree at
https://codeberg.org/pipcet/emacs-android (sorry about the repository
name, it's no longer accurate).  Note that there are "Actions" available
which have artifacts that you can download and test, if you are so
inclined.  Right now, for me, the ucrt64 build doesn't work but the
64-bit mingw32 build does.

If you're interested in testing this, please let me know which tree you
use so I can provide a patch that applies there and just contains the
setjmp fix.

>> Here is the patch I used for MPS though
>> https://raw.githubusercontent.com/kiennq/emacs-build/main/patches/mps/0002-mingw64-support-mingw64.patch
>
> Did you run the MPS test suite, and if you did, did all the tests
> succeed?

I must confess I haven't done that yet. As it has been decided to
deliberately break Emacs CI/CD using Wine, though, I'm not sure there's
any point.

> It is not recommended to try the MPS build of Emacs as long as the MPS
> library doesn't pass the tests with flying colors.  The patch to
> mpstd.h is the first place to look for possible issues.  E.g., why is
> MPS_PF_ALIGN 8 and not 16?

I don't believe we use that macro anywhere (scratch/igc uses 8 byte alignment on
all platforms) but yes, MPS_PF_ALIGN should be 16.

> Also, did you audit all the places
> conditioned on MPS_BUILD_MV and MPS_PF_W3I6MV and made sure they are
> conditioned on MPS_BUILD_GC and MPS_PF_W3I6GC when the MinGW64 build
> needs the same code as the 64-bit MSVC build?

I'm pretty sure I did, but I use a slightly different patch
(https://codeberg.org/pipcet/mps): Please be aware that this code does
run Emacs, but hasn't passed the test suite, so pretty much the only
thing you should be doing with it is to run the test suite.

Also, please keep in mind that -fno-omit-frame-pointer may be required
for GCC when optimizing. I spent entirely too much time tracking down a
bug that was ultimately due to that missing flag.

Pip