dropping 32-bit Windows host support

dropping 32-bit Windows host support

[Peter Maydell]

Our msys2 32-bit Windows host CI job has been failing recently
because upstream MSYS2 are starting to phase out 32-bit windows
host support and are steadily removing i686 versions of packages.
The latest is dtc:
https://gitlab.com/qemu-project/qemu/-/issues/2177

The writing is clearly on the wall for their 32-bit support, judging
from the "2023-12-13 - Starting to drop some 32-bit Packages" news
item at https://www.msys2.org/news/ and associated discussion at
https://github.com/msys2/MINGW-packages/discussions/19326 .

QEMU on a 32-bit host is not likely to be a great experience, and I
suspect we don't have many users using 32-bit Windows who couldn't
use the 64-bit version instead. Our Download page points users at
MSYS2's packages, and they have already dropped the 32-bit QEMU
package build. Stefan Weil's binaries page, which is the other thing
we link to from Download, has no 32-bit version newer than 7.2.0.
So anybody using 32-bit Windows QEMU must be building it themselves.
Plus, we've already deprecated system emulation on 32-bit x86 hosts,
so the only remaining "supported" config is with the NVMM or WHPX
accelerators.

I suggest that we should:

 * remove the msys2-32bit CI job entirely (we will still have at least
   compile-time coverage via the cross-win32-system job)
 * document that the 32-bit Windows support is deprecated in
   docs/about/build-platforms.rst and deprecated.rst
 * update our Download page to remove mention of 32-bit Windows

Any objections?

thanks
-- PMM

Re: dropping 32-bit Windows host support

[Daniel P. Berrangé]

On Mon, Feb 19, 2024 at 03:37:31PM +0000, Peter Maydell wrote:
> Our msys2 32-bit Windows host CI job has been failing recently
> because upstream MSYS2 are starting to phase out 32-bit windows
> host support and are steadily removing i686 versions of packages.
> The latest is dtc:
> https://gitlab.com/qemu-project/qemu/-/issues/2177
> 
> The writing is clearly on the wall for their 32-bit support, judging
> from the "2023-12-13 - Starting to drop some 32-bit Packages" news
> item at https://www.msys2.org/news/ and associated discussion at
> https://github.com/msys2/MINGW-packages/discussions/19326 .
> 
> QEMU on a 32-bit host is not likely to be a great experience, and I
> suspect we don't have many users using 32-bit Windows who couldn't
> use the 64-bit version instead. Our Download page points users at
> MSYS2's packages, and they have already dropped the 32-bit QEMU
> package build. Stefan Weil's binaries page, which is the other thing
> we link to from Download, has no 32-bit version newer than 7.2.0.
> So anybody using 32-bit Windows QEMU must be building it themselves.
> Plus, we've already deprecated system emulation on 32-bit x86 hosts,
> so the only remaining "supported" config is with the NVMM or WHPX
> accelerators.

The other data point is that Win11 is 64-bit only, and IIUC,
Win10 was 64-bit only for new OEM installs too, only upgrades
or end user installs could choose 32-bit. 

> 
> I suggest that we should:
> 
>  * remove the msys2-32bit CI job entirely (we will still have at least
>    compile-time coverage via the cross-win32-system job)
>  * document that the 32-bit Windows support is deprecated in
>    docs/about/build-platforms.rst and deprecated.rst
>  * update our Download page to remove mention of 32-bit Windows
> 
> Any objections?

I think that's sane.

With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

Re: dropping 32-bit Windows host support

[Thomas Huth]

On 19/02/2024 16.53, Daniel P. Berrangé wrote:
> On Mon, Feb 19, 2024 at 03:37:31PM +0000, Peter Maydell wrote:
>> Our msys2 32-bit Windows host CI job has been failing recently
>> because upstream MSYS2 are starting to phase out 32-bit windows
>> host support and are steadily removing i686 versions of packages.
>> The latest is dtc:
>> https://gitlab.com/qemu-project/qemu/-/issues/2177
>>
>> The writing is clearly on the wall for their 32-bit support, judging
>> from the "2023-12-13 - Starting to drop some 32-bit Packages" news
>> item at https://www.msys2.org/news/ and associated discussion at
>> https://github.com/msys2/MINGW-packages/discussions/19326 .
>>
>> QEMU on a 32-bit host is not likely to be a great experience, and I
>> suspect we don't have many users using 32-bit Windows who couldn't
>> use the 64-bit version instead. Our Download page points users at
>> MSYS2's packages, and they have already dropped the 32-bit QEMU
>> package build. Stefan Weil's binaries page, which is the other thing
>> we link to from Download, has no 32-bit version newer than 7.2.0.
>> So anybody using 32-bit Windows QEMU must be building it themselves.
>> Plus, we've already deprecated system emulation on 32-bit x86 hosts,
>> so the only remaining "supported" config is with the NVMM or WHPX
>> accelerators.
> 
> The other data point is that Win11 is 64-bit only, and IIUC,
> Win10 was 64-bit only for new OEM installs too, only upgrades
> or end user installs could choose 32-bit.

Yes, and considering that there will likely be a Windows 12 at one point in 
time, we'll drop support for Win10 and thus 32-bit support anyway.

>> I suggest that we should:
>>
>>   * remove the msys2-32bit CI job entirely (we will still have at least
>>     compile-time coverage via the cross-win32-system job)
>>   * document that the 32-bit Windows support is deprecated in
>>     docs/about/build-platforms.rst and deprecated.rst
>>   * update our Download page to remove mention of 32-bit Windows
>>
>> Any objections?
> 
> I think that's sane.

Sounds good to me, too.

Note that we already have an entry for "System emulation on 32-bit x86 
hosts" in the deprecation list ... so IMHO we could also justify to drop the 
32-bit Windows support immediately, since that's a subset of that entry.

  Thomas

Re: dropping 32-bit Windows host support

[Peter Maydell]

On Mon, 19 Feb 2024 at 16:26, Thomas Huth <thuth@redhat.com> wrote:
>
> On 19/02/2024 16.53, Daniel P. Berrangé wrote:
> > On Mon, Feb 19, 2024 at 03:37:31PM +0000, Peter Maydell wrote:
> >> Our msys2 32-bit Windows host CI job has been failing recently
> >> because upstream MSYS2 are starting to phase out 32-bit windows
> >> host support and are steadily removing i686 versions of packages.
> >> The latest is dtc:
> >> https://gitlab.com/qemu-project/qemu/-/issues/2177
> >>
> >> The writing is clearly on the wall for their 32-bit support, judging
> >> from the "2023-12-13 - Starting to drop some 32-bit Packages" news
> >> item at https://www.msys2.org/news/ and associated discussion at
> >> https://github.com/msys2/MINGW-packages/discussions/19326 .
> >>
> >> QEMU on a 32-bit host is not likely to be a great experience, and I
> >> suspect we don't have many users using 32-bit Windows who couldn't
> >> use the 64-bit version instead. Our Download page points users at
> >> MSYS2's packages, and they have already dropped the 32-bit QEMU
> >> package build. Stefan Weil's binaries page, which is the other thing
> >> we link to from Download, has no 32-bit version newer than 7.2.0.
> >> So anybody using 32-bit Windows QEMU must be building it themselves.
> >> Plus, we've already deprecated system emulation on 32-bit x86 hosts,
> >> so the only remaining "supported" config is with the NVMM or WHPX
> >> accelerators.
> >
> > The other data point is that Win11 is 64-bit only, and IIUC,
> > Win10 was 64-bit only for new OEM installs too, only upgrades
> > or end user installs could choose 32-bit.
>
> Yes, and considering that there will likely be a Windows 12 at one point in
> time, we'll drop support for Win10 and thus 32-bit support anyway.
>
> >> I suggest that we should:
> >>
> >>   * remove the msys2-32bit CI job entirely (we will still have at least
> >>     compile-time coverage via the cross-win32-system job)
> >>   * document that the 32-bit Windows support is deprecated in
> >>     docs/about/build-platforms.rst and deprecated.rst
> >>   * update our Download page to remove mention of 32-bit Windows
> >>
> >> Any objections?
> >
> > I think that's sane.
>
> Sounds good to me, too.
>
> Note that we already have an entry for "System emulation on 32-bit x86
> hosts" in the deprecation list ... so IMHO we could also justify to drop the
> 32-bit Windows support immediately, since that's a subset of that entry.

Mmm. I had initially misread that as only applying to TCG,
but it's more general than that. So I think I agree that we
should go ahead and mark 32-bit Windows support as definitively
dropped.

I'll put together some patches at some point.

-- PMM

Re: dropping 32-bit Windows host support

[Daniel P. Berrangé]

On Mon, Feb 19, 2024 at 04:40:38PM +0000, Peter Maydell wrote:
> On Mon, 19 Feb 2024 at 16:26, Thomas Huth <thuth@redhat.com> wrote:
> >
> > On 19/02/2024 16.53, Daniel P. Berrangé wrote:
> > > On Mon, Feb 19, 2024 at 03:37:31PM +0000, Peter Maydell wrote:
> > >> Our msys2 32-bit Windows host CI job has been failing recently
> > >> because upstream MSYS2 are starting to phase out 32-bit windows
> > >> host support and are steadily removing i686 versions of packages.
> > >> The latest is dtc:
> > >> https://gitlab.com/qemu-project/qemu/-/issues/2177
> > >>
> > >> The writing is clearly on the wall for their 32-bit support, judging
> > >> from the "2023-12-13 - Starting to drop some 32-bit Packages" news
> > >> item at https://www.msys2.org/news/ and associated discussion at
> > >> https://github.com/msys2/MINGW-packages/discussions/19326 .
> > >>
> > >> QEMU on a 32-bit host is not likely to be a great experience, and I
> > >> suspect we don't have many users using 32-bit Windows who couldn't
> > >> use the 64-bit version instead. Our Download page points users at
> > >> MSYS2's packages, and they have already dropped the 32-bit QEMU
> > >> package build. Stefan Weil's binaries page, which is the other thing
> > >> we link to from Download, has no 32-bit version newer than 7.2.0.
> > >> So anybody using 32-bit Windows QEMU must be building it themselves.
> > >> Plus, we've already deprecated system emulation on 32-bit x86 hosts,
> > >> so the only remaining "supported" config is with the NVMM or WHPX
> > >> accelerators.
> > >
> > > The other data point is that Win11 is 64-bit only, and IIUC,
> > > Win10 was 64-bit only for new OEM installs too, only upgrades
> > > or end user installs could choose 32-bit.
> >
> > Yes, and considering that there will likely be a Windows 12 at one point in
> > time, we'll drop support for Win10 and thus 32-bit support anyway.
> >
> > >> I suggest that we should:
> > >>
> > >>   * remove the msys2-32bit CI job entirely (we will still have at least
> > >>     compile-time coverage via the cross-win32-system job)
> > >>   * document that the 32-bit Windows support is deprecated in
> > >>     docs/about/build-platforms.rst and deprecated.rst
> > >>   * update our Download page to remove mention of 32-bit Windows
> > >>
> > >> Any objections?
> > >
> > > I think that's sane.
> >
> > Sounds good to me, too.
> >
> > Note that we already have an entry for "System emulation on 32-bit x86
> > hosts" in the deprecation list ... so IMHO we could also justify to drop the
> > 32-bit Windows support immediately, since that's a subset of that entry.
> 
> Mmm. I had initially misread that as only applying to TCG,
> but it's more general than that. So I think I agree that we
> should go ahead and mark 32-bit Windows support as definitively
> dropped.

Technically our existing deprecation would still allow for 32-bit
builds of qemu-img, qemu-io, etc tools, on Windows, but that's
very niche. 


With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

Re: dropping 32-bit Windows host support

[Stefan Weil via]

Am 19.02.24 um 17:26 schrieb Thomas Huth:

> On 19/02/2024 16.53, Daniel P. Berrangé wrote:
>> On Mon, Feb 19, 2024 at 03:37:31PM +0000, Peter Maydell wrote:
>>> Our msys2 32-bit Windows host CI job has been failing recently
>>> because upstream MSYS2 are starting to phase out 32-bit windows
>>> host support and are steadily removing i686 versions of packages.
>>> The latest is dtc:
>>> https://gitlab.com/qemu-project/qemu/-/issues/2177
>>> [...]


I agree with all your comments and also think that support for 32-bit 
Windows hosts can be dropped.

As Peter noted, I have been building 64-bit installers only since QEMU 
8.0.0. I have not received any complaints about this.

Best regards

Stefan