@dkwo you may want to try this.

In particular, I made ntl cross, could you make sure that cross built ntl works ok? We also need to adjust flintlib and singular templates so they are built with ntl on cross.

Amazing that you got ntl to cross build!
is it possible to merge that change (and the bumps of singular etc) as a separate pr?
this way i could try to build sage with those packages from void repos and test if it works.

This looks good to me.
@ahesford Perhaps this can be merged sometimes soon, so that at least sage is unbroken?

If this is ready to go, please clean up the commits (especially squashing the separate sagemath commits) and drop the "draft" badge.

@ahesford @dkwo this is now ready IMO.

Test failures are either transient (failed to download 'libpng-1.6.55_1' signature...) or some random failures in matrix_mod2_dense.pyx and matrix_integer_dense.pyx which have been around for a while and seem to get worse under load / CI.

@tornaria I have just rebuilt sage natively on aarch64, using cross ntl from void's repo: all tests pass and I can use sage in jupyter lab. Do you understand why the build is failing for x86_64* on the builders? it looks like it does not find maxima Program maxima found: NO and rebuilds it.

one difference is that on aarch64 maxima uses ecl.

also, on ci here Program maxima found: YES (/usr/bin/maxima)

Thanks!

The issue is that we need maxima-ecl on makedepends. When using -Q this is not a problem because maxima-ecl is on checkdepends.