allow gcc-15 from the system

[dimpase]

Fedora 42 comes with gcc-15 by default, so it should be allowed, as asked on sage-release

📝 Checklist

• The title is concise and informative.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes.
• I have updated the documentation and checked the documentation preview.

⌛ Dependencies

• linbox patches for gcc-15 and Apple clang-17 #39985 : linbox patches
• update givaro to 4.2.1 #39936 : givaro 4.2.1

[dimpase]

@tobiasdiez - CI still does useless Fedora 30, I don't know why. Can we replace it with Fedora 42 - this would test this PR, too

[enriqueartal]

The packages gap, singular and planarity have problems which can be solved using system package. I attarch the log for ecl.
ecl-24.5.10.log

[dimpase]

does Fedora have a usable ecl package?

I'll report this to upstream ecl, perhaps they already have a fix

[enriqueartal]

Not now. I have used a former src.rpm and I could produce a package using gcc-14. Then maxima produces also an error, I am trying to make a package maxima-runtime-ecl (again with gcc-14) which has been disabled in Fedora.

[enriqueartal]

With ecl and maxima system packages I get errors in sagelib,
sagelib-10.7.beta1.log
If I am not wrong they are related with linbox and ecl.

By the way, patch system package (2.8.1) is not accepted, no clue why in config.log

[dimpase]

Note that gcc-15 has default c23 C standard. This is a bump from c17. Adding -std=c17 to CFLAGS might help with ecl and maxima.

That is,

--- a/build/pkgs/ecl/spkg-install.in
+++ b/build/pkgs/ecl/spkg-install.in
@@ -1,6 +1,6 @@
 cd src
 
-export CFLAGS
+export CFLAGS="-std=c17 $CFLAGS"
 export CXXFLAGS
 export LDFLAGS

(and maxima gets its flags from ecl).

[dimpase]

By the way, patch system package (2.8.1) is not accepted, no clue why in config.log
most probably it reports its version as 2.8, that's why. Anyway, this is fixed by #39943
(testing appreciated)

[github-actions]

Documentation preview for this PR (built with commit 68c24f8; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[enriqueartal]

No luck with sagelib with or without system-packages. On the other side, patch is no longer a package using #39943

[dimpase]

By the way, patch system package (2.8.1) is not accepted, no clue why in config.log
most probably it reports its version as 2.8, that's why. Anyway, this is fixed by #39943
(testing appreciated)

[dimpase]

ECL problem reported here:
https://gitlab.com/embeddable-common-lisp/ecl/-/issues/775

[dimpase]

No luck with sagelib with or without system-packages.

please post errors. Are they again due to c23 used for C, or there are also C++ issues?

[enriqueartal]

Thanks. I am not sure how to proceed. First I modified spkg-install.in as in #39977 (comment). I attach sagelib-10.7.beta1.log using ecl and system packages. Then I remove ecl system package, and this is the log of failure: ecl-24.5.10.log. Finally, I tried to compile maxima with ecl system package: maxima-5.47.0.log.
I hope it may help.

[tobiasdiez]

@tobiasdiez - CI still does useless Fedora 30, I don't know why. Can we replace it with Fedora 42 - this would test this PR, too

Sure, just edit

sage/.github/workflows/ci-linux-incremental.yml

Line 102 in e0cf1e4

"fedora-40",]

and the ci-linux.

[dimpase]

@enriqueartal : in regard of ecl, can you replace c17 in CFLAGS with gnu17

i.e. in the patch above it should be

-export CFLAGS
+export CFLAGS="-std=gnu17 $CFLAGS" 

[enriqueartal]

One step achieved. Now ecl and maxima are OK; still issues with sagelib regarding ecl and linbox.

[dimpase]

@tobiasdiez - CI still does useless Fedora 30, I don't know why. Can we replace it with Fedora 42 - this would test this PR, too

Sure, just edit

sage/.github/workflows/ci-linux-incremental.yml

Line 102 in e0cf1e4

"fedora-40",]

and the ci-linux.

How about doing this on top of your recent PR updating CI? (I meant #39942)

[dimpase]

One step achieved. Now ecl and maxima are OK; still issues with sagelib regarding ecl and linbox.

I see issues with linbox (more patches needed)

With ecl, running plane C (gcc) while building sage.matrix.matrix_numpy_integer_dense extension, and anything else which includes ecl.h, should be done with -std=gnu17.
Can you try something like

CFLAGS="-std=gnu17" make -j8 build

@tobiasdiez - what to do here in meson.build? Does it have something line extension_data_c, analogous to extension_data_cpp ? (I am still unable to use meson docs - e.g. I cannot find anything in extension_data_cpp)

[enriqueartal]

I think ecl's issues are gone, still issues with linbox with sagelib
sagelib-10.7.beta1.log

[dimpase]

Do you use system's givaro? Or a built one? Anyway, please try #39936 (for givaro - but it's important, as linbox depends on givaro)

[enriqueartal]

I use system's givaro (4.2.1). I can try with spkg.

[dimpase]

I use system's givaro (4.2.1). I can try with spkg.

It's fine, should make no difference. For linbox, I think, 2 more patches are needed (incidentally they are needed for the new Apple's clang 17), as I added here: https://github.com/Macaulay2/homebrew-tap/blob/main/Formula/linbox.rb

Specifically, add to build/pkgs/linbox/patches "https://github.com/linbox-team/linbox/commit/d1f618fb0ee4a84be3ccddcfc28b257f34e1cbf7.patch?full_index=1"
and "https://github.com/linbox-team/linbox/commit/4a1e1395804d4630ec556c61ba3f2cb67e140248.patch?full_index=1"
(save them as 41.patch and 42.patch, so that they apply in the order following the rest of the patches)

[enriqueartal]

Ok for linbox spkg but sagelib not OK

[dimpase]

I gather linbox needs

--- a/linbox/blackbox/block-hankel.h
+++ b/linbox/blackbox/block-hankel.h
@@ -345,8 +345,8 @@ namespace LinBox
 		template<class Vector1, class Vector2>
 		Vector1& apply(Vector1 &x, const Vector2 &y) const
 		{
-			linbox_check(this->_coldim == y.size());
-			linbox_check(this->_rowdim == x.size());
+			linbox_check(this->coldim() == y.size());
+			linbox_check(this->rowdim() == x.size());
 			BlasMatrixDomain<Field> BMD(field());
 #ifdef BHANKEL_TIMER
 			_chrono.clear();

as well

[dimpase]

and probably more, from linbox PRs 320, 322

[enriqueartal]

Thanks! I saved as 43.patch and now it works. If my memory is OK gap and singular had also issues with gcc-15; for gap, even outside sage.

[enriqueartal]

Last thing for today: gap spkg is OK, not for singular. Thanks.

[enriqueartal]

and probably more, from linbox PRs 320, 322

I used nothing from these PRs

[dimpase]

I've put the linbox patches on #39985
Please update your branch accodingly

[enriqueartal]

Another test in a virtual machine failed with contourpy. In the former installation, it is ok for contourpy.

[dimpase]

somewhere the sin of passing CFLAGS to a C++ compiler is committed

[enriqueartal]

somewhere the sin of passing CFLAGS to a C++ compiler is committed

It seems that for some packages

CFLAGS="-std=gnu17" make -jn build

is needed and for others it is just make -jn build. Somehow I has to combine them, including maybe some intermediate cleaning for sagelib.