Cleanup sage.env

[Snark]

1. sage.all should be importable and mostly usable even if no environment variables like SAGE_ROOT or SAGE_LOCAL are set.

2. Some unneeded variables are removed.

3. Get rid of the over-engineered $VAR replacement stuff.

4. In some places, SAGE_LIB (installed location of Sage) can be used instead of SAGE_SRC.

5. Determine SAGE_LIB not from site-packages but from sage.__file__

There is a ​thread on sage-devel that complains about the failure in Debian of import sage.all from system Python.

[fbissey]

SAGE_ROOT is harder. It doesn't correspond to any configuration features of anything. At best it is SAGE_LOCAL/... Any corners in which that wouldn't be true?

[dunfield]

Replying to fbissey:

SAGE_ROOT is harder. It doesn't correspond to any configuration features of anything. At best it is SAGE_LOCAL/... Any corners in which that wouldn't be true?

One place where the SAGE_LOCAL/.. pattern does not hold is the Debian package, where SAGE_LOCAL=/usr but SAGE_ROOT=/usr/share/sagemath. Moreover, on Debian SAGE_SCRIPTS_DIR is $SAGE_ROOT/bin rather than the more typical $SAGE_LOCAL/bin.

[fbissey]

I see. debian choose to do something like that. In sage-on-gentoo SAGE_ROOT points to a location where there used to be something but is now empty. Does debian use SAGE_ROOT anywhere else? Otherwise they could just define SAGE_SCRIPTS_DIR directly.

[dunfield]

Replying to fbissey:

Does debian use SAGE_ROOT anywhere else? Otherwise they could just define SAGE_SCRIPTS_DIR directly.

To clarify, Debian doesn't change anything in env.py related to SAGE_SCRIPTS_DIR; the only fallback there is $SAGE_LOCAL/bin which does not contain the relevant scripts. Instead, in /usr/share/sagemath/bin/sage-env they manually set SAGE_ROOT=/usr/share/sagemath/ and SAGE_SCRIPTS_DIR=/usr/share/sagemath/bin.

On Debian, I tried setting SAGE_ROOT to a garbage value in sage-env, and Sage still started and worked fine near as I can tell.

[fbissey]

I can see it now. In vanilla sage SAGE_SCRIPTS_DIR is only set in sage-env and used to source sage-env-config. But I certainly know that debian has plan for that folder.

So as far as we can see, it would be fine to set SAGE_ROOT in env.py. But it could probably be left as is without impacting anything and doing from sage import all in python should work without setting it.

Looking at it there is one thing I have set SAGE_ROOT for! sage/misc/copying.py tries to access SAGE_ROOT/COPYING.txt. So its current value in sage-on-gentoo is just so that bit works :) I need to change that to be more conformant to Gentoo policies.

There is a somewhat harmless appearance of SAGE_ROOT in sage/misc/citation.pyx. There is a more disturbing one in sage/misc/persist.pyx where I would have used DOT_SAGE to be honest. The rest is mostly doctesting, packaging, doctests themselves and documentation strings.

[fbissey]

Scope extension. That's brave! I notice you mention SAGE_SRC and SAGE_LIB, is it because you spotted I have

_add_variable_or_fallback('SAGE_SRC',        opj('$SAGE_LIB'))

in sage-on-gentoo? I override it with the appropriate value during build of course. But it should work in most places.

[jdemeyer]

Replying to fbissey:

is it because you spotted I have

_add_variable_or_fallback('SAGE_SRC',        opj('$SAGE_LIB'))

in sage-on-gentoo?

Actually no. It's something I figured out independently.

[fbissey]

Will using sage.__file__ change the way we use SAGE_LIB or do you plan to restore the value starting from it?