Incompatibilities in Sage mpmath cause downstream issues in SymPy

[asmeurer]

The Sage mpmath has some incompatibilities with non-sage mpmath, which causes some issues in SymPy, since mpmath always uses the Sage backend if Sage is installed.

You can get a full idea of the issues by running the SymPy tests in Sage (import sympy;sympy.test()). A few failures are also due to some gmpy2 issues (gmpy2 types fail on Sage integers).

The issues I've noticed so far are:

• The sage mpmath types do not subclass from mpmath.ctx_mp.mpnumeric, which is the base class of mpf and mpc according to its own docstring.

• The sage mpf type does not have a context attribute.

Another issue, reported in sympy/sympy#19690:

import sympy.physics
from sympy.physics.units import meter
import copy
copy.deepcopy(3.7*meter)
[...]
File "sage/rings/integer.pyx", line 7143, in sage.rings.integer.mpz_set_str_python (build/cythonized/sage/rings/inte
ger.c:43554)
TypeError: unable to convert '0xecccccccccccd' to an integer

(fixed in sympy/sympy#21996, merged for inclusion in Sympy 1.9)

References:

• Install Sage on Travis using conda-forge sympy/sympy#14703
• doctesting sage-9.3.beta failures cschwan/sage-on-gentoo#628
• Enable sage backend by default only if SAGE_ROOT is set mpmath/mpmath#466
• sage-devel, 2021-03, pointy python question related to sage
• sage-devel, 2021-03, Running MathicsSession inside Sage
• Explicitly set MPMATH_SAGE env variable #32799: Explicitly set MPMATH_SAGE env variable

CC: @antonio-rojas @kiwifb @fredrik-johansson @isuruf @nbruin @saraedum @slel @soehms @strogdon @videlec

Component: distribution

Keywords: mpmath

Issue created by migration from https://trac.sagemath.org/ticket/25445

[slel]

Changed keywords from none to mpmath

[slel]

Description changed:

--- 
+++ 
@@ -4,9 +4,14 @@
 
 The issues I've noticed so far are:
 
-- The sage mpmath types do not subclass from mpmath.ctx_mp.mpnumeric, which is the base class of mpf and mpc according to its own docstring.
+- The sage mpmath types do not subclass from `mpmath.ctx_mp.mpnumeric`, which is the base class of `mpf` and `mpc `according to its own docstring.
 
-- The sage mpf type does not have a context attribute. 
+- The sage `mpf` type does not have a context attribute. 
 
 
-XREF: https://github.com/sympy/sympy/pull/14703
+References:
+
+- https://github.com/sympy/sympy/pull/14703
+- https://github.com/cschwan/sage-on-gentoo/issues/628
+- https://github.com/fredrik-johansson/mpmath/pull/466
+- [sage-devel, 2021-03, pointy python question related to sage](https://groups.google.com/g/sage-devel/c/MSDXHtzb00Q)

[kiwifb]

comment:2

The content of the description echoes some of what I said on sage-devel, we are probably in this situation because sage misuses mpmath, or uses some parts that should be private.

[soehms]

Description changed:

--- 
+++ 
@@ -15,3 +15,4 @@
 - https://github.com/cschwan/sage-on-gentoo/issues/628
 - https://github.com/fredrik-johansson/mpmath/pull/466
 - [sage-devel, 2021-03, pointy python question related to sage](https://groups.google.com/g/sage-devel/c/MSDXHtzb00Q)
+- [sage-devel, 2021-03, Running MathicsSession inside Sage](https://groups.google.com/g/sage-devel/c/hUJeK4r34k8)

[dimpase]

comment:7

mpmath update on #31564