[ITA] llvm

[ITA] llvm

[徐持恒 Xu Chiheng]

I'd like to adopt the llvm package.
I have build scripts and patches, and .cygport files at
https://github.com/xu-chiheng/Note.
16.0.6 17.0.6 and 18.0.0(current) are supported.

Re: [ITA] llvm

[ASSI]

徐持恒 Xu Chiheng via Cygwin-apps writes:
> I'd like to adopt the llvm package.
> I have build scripts and patches, and .cygport files at
> https://github.com/xu-chiheng/Note.

That's mildly confusing to me…  anyway, I'll try to build LLVM from this
(maybe this weekend).  I expect there will need to be further changes to
the packaging, given how far out of date the cygport you based it on is.

> 16.0.6 17.0.6 and 18.0.0(current) are supported.

Did you run the test suite?


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Factory and User Sound Singles for Waldorf Blofeld:
http://Synth.Stromeko.net/Downloads.html#WaldorfSounds

Re: [ITA] llvm

[徐持恒 Xu Chiheng]

On Thu, Jan 11, 2024 at 1:04 AM ASSI via Cygwin-apps
<cygwin-apps@cygwin.com> wrote:
>
> That's mildly confusing to me…  anyway, I'll try to build LLVM from this
> (maybe this weekend).  I expect there will need to be further changes to
> the packaging, given how far out of date the cygport you based it on is.
>
The HEAD revision 601443d517a84e2eceee5eceb0c6bdae76adeda8 2024-01-10

> > 16.0.6 17.0.6 and 18.0.0(current) are supported.
>
> Did you run the test suite?
>
Not run the test suite. But I have bubble tests, including
bootstrapping Clang, building GCC and Binutils.

Re: [ITA] llvm

[徐持恒 Xu Chiheng]

https://github.com/xu-chiheng/Note/blob/main/Tool/llvm.cygport

Re: [ITA] llvm

[ASSI]

ASSI via Cygwin-apps writes:
> That's mildly confusing to me…  anyway, I'll try to build LLVM from this
> (maybe this weekend).  I expect there will need to be further changes to
> the packaging, given how far out of date the cygport you based it on is.

LLVM 16 does indeed build (didn't look at the other two versions), but
the cygport misses all the work to actually produce packaging that can
be used to upgrade from the existing Cygwin version.  Source file is he
whole project instead of just LLVM + Polly and the build also enables
clang (which should be in a separate package).  Generator has been
changed from Ninja to Make (and switching back to Ninja makes the
install step fail because that target wasn't made available).  Multiple
build options have been changed from what they were in Cygwin, I have
not had time to investigate what that actually means for the resulting
compiler binaries.  I don't understand why the
process_install_dir_bin_symlinks function is necessary, as cygport
should already fix up symlinks in the install dir.  I did enable the
test suite and it tests with 259 fails (was 180 fails before and looks
about the same general pattern as with version 9.0.1).


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Factory and User Sound Singles for Waldorf Blofeld:
http://Synth.Stromeko.net/Downloads.html#WaldorfSounds

Re: [ITA] llvm

[徐持恒 Xu Chiheng]

On Mon, Jan 15, 2024 at 4:51 AM ASSI via Cygwin-apps
<cygwin-apps@cygwin.com> wrote:
>
> LLVM 16 does indeed build (didn't look at the other two versions), but
> the cygport misses all the work to actually produce packaging that can
> be used to upgrade from the existing Cygwin version.  Source file is he
> whole project instead of just LLVM + Polly and the build also enables
> clang (which should be in a separate package).  Generator has been
> changed from Ninja to Make (and switching back to Ninja makes the
> install step fail because that target wasn't made available).  Multiple
> build options have been changed from what they were in Cygwin, I have
> not had time to investigate what that actually means for the resulting
> compiler binaries.  I don't understand why the
> process_install_dir_bin_symlinks function is necessary, as cygport
> should already fix up symlinks in the install dir.  I did enable the
> test suite and it tests with 259 fails (was 180 fails before and looks
> about the same general pattern as with version 9.0.1).
>

Without process_install_dir_bin_symlinks, you can't get an installable
package *.xz.
For simplicity, I personally think to put all llvm related packages in
an all-in-one "llvm" package is a good idea.
The new llvm package is going to replace the old clang and llvm, and
maybe other llvm related packages.

Re: [ITA] llvm

[ASSI]

徐持恒 Xu Chiheng via Cygwin-apps writes:
> Without process_install_dir_bin_symlinks, you can't get an installable
> package *.xz.

You have not explained what exactly the issue is you're trying to solve.
If you get links to outside DESTDIR then the install targets are broken
and you need to fix the Makefiles.  If something else, please explain.
Again, there already is a pass in cygport that's supposed to fix up
symlinks, if that's not working we need to figure out why and how to fix
it if necessary.

> For simplicity, I personally think to put all llvm related packages in
> an all-in-one "llvm" package is a good idea.

I have sympathy for any idea that makes the life of the maintainer
easier…

> The new llvm package is going to replace the old clang and llvm, and
> maybe other llvm related packages.

…but ignoring the needs of the user isn't the solution to this problem.
Forcing everyone to install everything just because they want to use
something that links a library related to LLVM?  Please not.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

SD adaptations for KORG EX-800 and Poly-800MkII V0.9:
http://Synth.Stromeko.net/Downloads.html#KorgSDada

Re: [ITA] llvm

[徐持恒 Xu Chiheng]

On Tue, Jan 16, 2024 at 2:53 PM ASSI via Cygwin-apps
<cygwin-apps@cygwin.com> wrote:
>
> 徐持恒 Xu Chiheng via Cygwin-apps writes:
> > Without process_install_dir_bin_symlinks, you can't get an installable
> > package *.xz.
>
> You have not explained what exactly the issue is you're trying to solve.
> If you get links to outside DESTDIR then the install targets are broken
> and you need to fix the Makefiles.  If something else, please explain.
> Again, there already is a pass in cygport that's supposed to fix up
> symlinks, if that's not working we need to figure out why and how to fix
> it if necessary.
>

The symlink DESTDIR/usr/clang++ was wrongly processed by cygport, the
resulting bin/clang++ can't be executed, but /bin/clang is ok.
This caused clang can't bootstrap. I don't know why, it is clearly a
cgyport bug, I just remove the symlink, and copy it.