Problem building sagelib-10.10.beta1

[Simon King]

Dear developers,

again, after some ubuntu upgrade, I have to rebuild sagemath (somehow I miss the old times when sage did NOT use system packages) and encounter problems. The error says ERROR: Unknown option: "defer_feature_checks".

See attached logs. What can I do to get sagemath to build?

Cheers
Simon

[seb....@gmail.com]

Hi Simon,

this error currently also appears in many CI logs:

* Build & Test: https://github.com/sagemath/sage/actions/runs/26059207276/job/76614820771#step:9:2286
* Build Docker images ...: https://github.com/sagemath/sage/actions/runs/26059203196/job/76614808589#step:8:22241

It seem that something general is broken in 10.10.beta1.

Sebastian

[Simon King]

Hi Sebastian,

thank you. So, I hope that the git master branch will work (I just checked it out).

Best regards
Simon

[Michael Orlitzky]

You have to clean (delete) the sagelib build directory to force a
reconfigure:

$ rm -rf build/sage-distro

defer_feature_checks is a new option in meson.options that the sage
library makes use of, but the usual rebuild login in meson doesn't
seem to handle the addition of a new option without a fresh build.

I added the rm -rf for the benefit of the CI, but removed it from the
final PR because starting from scratch would slow down every
subsequent rebuild:

https://github.com/sagemath/sage/compare/e6cff7577af771a93a3f24bd477aff6ff1c0b5be..a240736dd08f1fa78953064ba064cd52a2a23a2d

Apologies for the headache. I've been meaning to report the issue
upstream to meson.

[Michael Orlitzky]

On 2026-05-19 00:11:47, seb....@gmail.com wrote:
> Hi Simon,
>
> this error currently also appears in many CI logs:
>

I think I could add back the rm -rf with a "CI fix" label but then not
merge the PR?

[Michael Orlitzky]

On 2026-05-19 07:17:25, Michael Orlitzky wrote:
>
> I think I could add back the rm -rf with a "CI fix" label but then not
> merge the PR?

https://github.com/sagemath/sage/pull/42209

[Nils Bruin]

On Monday, 18 May 2026 at 22:42:06 UTC-7 Simon King wrote:

(somehow I miss the old times when sage did NOT use system packages)

Conda-based installs give you something that is virtually independent of system packages. It does that by replacing "system" by "conda", but one generally has much more freedom/control over when one updates "conda" packages.

[Dima Pasechnik]

Currently  starting to test Linuxbrew build for Sage (Linuxbrew is basically Homebrew, but on Linux, using the same formulae (i.e. installation scripts written in Ruby); Macaulay2 builds on Linuxbrew, and shares quite a few deps with SageMath).  https://docs.brew.sh/Homebrew-on-Linux

As we are basically able to do Meson build using Homebrew packages, I hope Linuxbrew can be used in a similar way on mainstream  Linux systems (Fedora/Ubuntu/Debian) which don't provide usable non-Python dependencies. It's slightly less convenient to install, as it wants to be installed in /home/linuxbrew, but on the other hand it's not as centralised as conda:

one can create one's own formulae with binary installs, see for example https://github.com/dimpase/homebrew-tap

Then sage-distro can be retired. :-)

Dima

[Michael Orlitzky]

A non-editable install will be a lot better with regard to
rebuilds. In the absence of any other information, meson generates
dependencies on e.g. /usr/lib/libgmp.so in the ninja rules. If your
distro updates gmp, even in a 100% compatible way, an editable install
will rebuild sage the next time you start it.

Non-editable installs OTOH just sit there on your disk. They will
eventually break, but only when a library update changes or removes
something that sage actually uses. When that happens, you rebuild
manually. If you're on a stable distro that should be rare as is IMO
the more user-friendly approach, especially for developers.