Upstream {fmpz,fmpq}_poly_extra

[saraedum]

Almost everything should be upstreamed to FLINT.

There is no problem if for some time FLINT and e-antic ship the same symbols here (as long as they are compatible.)

[w-bruns]

Unfortunately I ran into a problem when I tried to install ea-ntic 2.0.2 for Normaliz under MSYS2 for MS Windows:

C:/msys64/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/14.2.0/../../../../x86_64-w64-mingw32/bin/ld.exe: C:/Users/wbruns/Dropbox/NormalizMSYS/New/normaliz-3.10.4/local/lib/libflint.a(fmpz_poly_merged.o): in function `fmpz_poly_randtest_irreducible':
C:\Users\wbruns\Dropbox\NormalizMSYS\New\normaliz-3.10.4\nmz_opt_lib\Flint_source\flint-3.0.1/src/fmpz_poly/randtest.c:89: multiple definition of `fmpz_poly_randtest_irreducible'; fmpz_poly_extra/.libs/randtest_irreducible.o:C:\Users\wbruns\Dropbox\NormalizMSYS\New\normaliz-3.10.4\nmz_opt_lib\E-ANTIC_source\e-antic-2.0.2\libeantic\src/fmpz_poly_extra/randtest_irreducible.c:18: first defined here

I have not experienced this problem under Linux and MacOS. But it would be good if it could be removed completely.

[saraedum]

fmpz_poly_randtest_irreducible is only available since c7e042fdafa0a25db7f23ef2b7601e01da43b82f in FLINT which I think means FLINT 3. So, if we remove it completely, we'd have to drop support for FLINT 2.

[saraedum]

Can you share the output of the configure run? There was apparently an error when trying to detect whether these symbols are available.

[w-bruns]

Let me repeat that the problem only arises under MSYS2. I am attaching config.status from the last brun, which was successful since I have patched the critical file in e-antic.
config.zip

[saraedum]

Could I get the config.log of the failed run?

[w-bruns]

config.log

It could be that this olog belongs to a build with the patched e-antic. But I hope the patch had no influence on configure. If necessary I can reproduce the failure.

[saraedum]

Have you tried to set LDFLAGS="$LDFLAGS --allow-multiple-definition"? The linkers on macOS and Linux do not complain about a symbol being repeated.

[saraedum]

At the moment, we have no CI setup for msys2 here so it is annoying to test this. Related is also conda-forge/e-antic-feedstock#24.

[w-bruns]

I have not tried LDFLAGS="$LDFLAGS --allow-multiple-definition", but will do so.

[w-bruns]

As I have just seen, I can avoid the problem by installing e-antic using only Flint as insatlled via pacman. Then there is no libflint.a around, but e-antic nevertheless builds a static libe-antic.a. For normaliz.exe I need libflint.a and I can install it after e-antic has been made. Then everything is fine.

Nevertheless, under Linux or MacOS a libflint.a causes no problem. Therefore I will test LDFLAGS="$LDFLAGS --allow-multiple-definition.

[w-bruns]

Does not seem to help. So I stay with my patch. It is a more stable solution, even if it can be avoided by another order of installation steps.

[saraedum]

Ok. Static builds are weird (but somewhat more standard on Windows.) If the problem shows up in #276, then I'll put a fix in, otherwise, it's probably easiest to stick with your patch.