Explicitly set MPMATH_SAGE env variable

[antonio-rojas]

Since mpmath 1.2.1, specifically since commit [1], mpmath checks for the SAGE_ROOT or MPMATH_SAGE env variables in order to enable sage types instead of relying on whether sage.all can be imported.

This breaks distro packages where SAGE_ROOT is undefined. I reported this upstream a while ago [2] with not much luck. Here we explicitly define MPMATH_SAGE in sage-env so mpmath uses Sage types in a Sage session even if SAGE_ROOT is not defined.

This is still broken when using sage as a python library (in which case sage-env is not read at all), but fixing that use case is beyond the scope of this ticket.

[1] mpmath/mpmath@84ca137
[2] mpmath/mpmath#574

CC: @mkoeppe @kiwifb @dimpase @isuruf

Component: packages: standard

Author: Antonio Rojas

Branch/Commit: d4b1b60

Reviewer: François Bissey

Issue created by migration from https://trac.sagemath.org/ticket/32799

[antonio-rojas]

Branch: u/arojas/explicitly_set_mpmath_sage_env_variable

[antonio-rojas]

Author: Antonio Rojas

[antonio-rojas]

New commits:

c2cde67

Explicitly set MPMATH_SAGE env variable

[antonio-rojas]

Commit: c2cde67

[slel]

comment:3

See also

• Incompatibilities in Sage mpmath cause downstream issues in SymPy #25445: Incompatibilities in Sage mpmath cause downstream issues in SymPy
• sage-devel, 2021-10-17:
  BUG: Si(pi).n() throws error in sage 9.2-2
• sage-devel, 2021-03-18:
  Pointy Python question related to Sage

[kiwifb]

comment:4

This is the minimum to do vanilla sage. I have already endorsed the approach in principle months ago. I am not sure I would check MPMATH_SAGE first since I'd always want it to be 1, unless I explicitly want to break things.

I would like more opinion on that one because not providing a way to override is contentious.

[mkoeppe]

comment:5

We should probably set this environment variable in the Sage Python library, not in the sage-env script.

[kiwifb]

comment:6

Replying to @mkoeppe:

We should probably set this environment variable in the Sage Python library, not in the sage-env script.

I agree with that. If I knew how to do it, there would already be a ticket for it ready to review. The stuff I tried didn't work, the only things I can think of require cooperation from mpmath devs.

[sagetrac-git]

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

d4b1b60

Set the env variable in the Sage python library

[sagetrac-git]

Changed commit from c2cde67 to d4b1b60

[antonio-rojas]

comment:8

Replying to @kiwifb:

I agree with that. If I knew how to do it, there would already be a ticket for it ready to review. The stuff I tried didn't work, the only things I can think of require cooperation from mpmath devs.

Does this not work for you?

[kiwifb]

comment:9

No, because it is too late to set an environment variable. You are already in the application and it is run with the variables it was launched with. You would need a python variable to tell mpmath that it has been imported from sage.

[kiwifb]

comment:10

It would work if you started a new application like what happens in sage/interface/*.py with pexpect but not while running.

[antonio-rojas]

comment:11

Replying to @kiwifb:

No, because it is too late to set an environment variable. You are already in the application and it is run with the variables it was launched with. You would need a python variable to tell mpmath that it has been imported from sage.

Can you give more details on how exactly it is failing for you? Because everything seems to work fine for me (both doctests and running stuff from a sage session)

[kiwifb]

Reviewer: François Bissey

[kiwifb]

comment:12

OK, it surprisingly works. I must have had a messed up environment when I tried it.

[vbraun]

Changed branch from u/arojas/explicitly_set_mpmath_sage_env_variable to d4b1b60