Make arb config more flexible and find also flint-arb

[tobiasdiez]

If one has flint-arb installed but not arb as a system package, then compilation currently fails unless one explicitly sets ARB_LIBRARY = flint-arb. In this ticket, we also check for the existence of flint-arb making this manual step obsolete. This is similar in spirit to #30706, which does the same for CBLAS.

The config file build/pgks/arb/spkg-configure.m4 does a similar check (I think) and maybe can/needs to be adapted as well. If that's the case I would like to ask if somebody else can implement the necessary changes, as I've no experience with these config files. Thanks!

Depends on #30901

CC: @mkoeppe

Component: build

Branch/Commit: public/build/arbconfig @ 5d606e5

Issue created by migration from https://trac.sagemath.org/ticket/30936

[tobiasdiez]

comment:2

Merged the latest develop branch and version of #30901.

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Branch pushed to git repo; I updated commit sha1. New commits:

0e31204	Merge branch 'develop' of git://github.com/sagemath/sage into public/build/arbconfig
3fcaf5f	Merge branch 'develop' of git://github.com/sagemath/sage into public/build/multiarchsimple
090e6f1	Simplify code
fa4556a	Remove _get_sage_local
5d606e5	Merge branch 'public/build/multiarchsimple' of git://trac.sagemath.org/sage into public/build/arbconfig

[7ed8c4ca-6d56-4ae9-953a-41e42b4ed313]

Changed commit from c54df32 to 5d606e5

[mkoeppe]

comment:4

Debian renames libarb to libflint-arb because there is already a different library taking the name libarb. So code that only checks the presence of libarb or libflint-arb (in whatever order) cannot be correct.

[tobiasdiez]

comment:5

The code only provides a reasonable fall-back if SAGE_ARB_LIBRARY doesn't exist. Thus users on debian just have to set SAGE_ARB_LIBRARY = "flint-arb" and everything works as before.
Do I miss something?

(The purpose of this ticket was not to reimplement the checks in arb/spkg-configure in the setup.py file.)

[mkoeppe]

comment:6

The ticket tries to support a use case that is not currently supported (installation of sagelib without running ./configure). Instead of putting more complicated defaults into env.py, it would be better to develop a proper solution in sage.features.

[tobiasdiez]

comment:7

Want I want to support is installation of sagelib without the sage_conf package since I don't know how to install the latter with pipenv (I could install it by specifying the relative path in the pipfile, but that really ugly). It would be more practical if sage_conf would be a config file in src instead of a python package, but we discussed this already elsewhere...

So in this ticket I provided a reasonable fallback, similar to what is done for singular and gap. How would this be implemented with sage.features?

[mkoeppe]

comment:8

Replying to @tobiasdiez:

Want I want to support is installation of sagelib without the sage_conf package since I don't know how to install the latter with pipenv (I could install it by specifying the relative path in the pipfile, but that really ugly).

The Sage distribution builds a wheel of it, you would just install from there.

[mkoeppe]

comment:9

Replying to @tobiasdiez:

How would this be implemented with sage.features?

You would write a class similar to CythonFeature (from src/sage/features/__init__.py) with a method revealing the library name. The class would use the information from sage.env.ARB_LIBRARY (get rid of the default); but if unset, it would try to link a test program using acb_mat_eig_simple against the library. In fact, using CythonFeature for doing this test could pretty much work -- except that I'm not sure if sage.misc.cython perhaps pulls in too many modules at the moment.

[tobiasdiez]

Changed author from Tobias Diez to none

[tobiasdiez]

comment:10

That sounds reasonable. But it's also more work than I planned to invest into this, so I'll leave this to somebody else to implement.

[mkoeppe]

Outdated; we have SAGE_ARB_LIBRARY.