This breaks sagemath (which sets MPMATH_SAGE=1 in the environment and expects the mantissa to be of type sage.rings.integer.Integer. I'm testing with mpmath 1.4.0a0, now that sympy 1.12.1a1 supports it.

Is there a plan on how to move forward sagemath?

cc: @dimpase

@tornaria - I don't recall what the plan was - perhaps just to drop that Sage variable?

@tornaria - I don't recall what the plan was - perhaps just to drop that Sage variable?

First sagemath has to support mpmath with gmpy2 backend, but I did not find a PR for that (maybe there is none, I just saw this being discussed on #705)

@skirpichev @fredrik-johansson do you have any input about this?

FWIW, I think removing the sage backend is a good move, but some coordination with sagemath seems a good thing.

First sagemath has to support mpmath with gmpy2 backend

I don't think we should do anything special for it. If sage has the gmpy2 module - gmpy2 support for the mpmath will be turned on, unless you disabled it by the MPMATH_NOGMPY environment variable.

First sagemath has to support mpmath with gmpy2 backend

I don't think we should do anything special for it. If sage has the gmpy2 module - gmpy2 support for the mpmath will be turned on, unless you disabled it by the MPMATH_NOGMPY environment variable.

Something has to be done somewhere, since running sagemath with 1.4.0a0 is broken, for instance:

$ sage
┌────────────────────────────────────────────────────────────────────┐
│ SageMath version 10.3, Release Date: 2024-03-19                    │
│ Using Python 3.12.2. Type "help()" for help.                       │
└────────────────────────────────────────────────────────────────────┘
sage: from sage.libs.mpmath.all import sage_to_mpmath, mpmath_to_sage
sage: mpmath_to_sage(sage_to_mpmath(0.0, prec=50), prec=50)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[2], line 1
----> 1 mpmath_to_sage(sage_to_mpmath(RealNumber('0.0'), prec=Integer(50)), prec=Integer(50))

File /usr/lib/python3.12/site-packages/sage/libs/mpmath/utils.pyx:271, in sage.libs.mpmath.utils.mpmath_to_sage (build/cythonized/sage/libs/mpmath/utils.c:8440)()
    269 if hasattr(x, "_mpf_"):
    270     y = RealField(prec)()
--> 271     mpfr_from_mpfval(y.value, x._mpf_)
    272     return y
    273 elif hasattr(x, "_mpc_"):

File /usr/lib/python3.12/site-packages/sage/libs/mpmath/utils.pyx:166, in sage.libs.mpmath.utils.mpfr_from_mpfval (build/cythonized/sage/libs/mpmath/utils.c:7571)()
    164 cdef long exp
    165 cdef long bc
--> 166 sign, man, exp, bc = x
    167 if man:
    168     mpfr_set_z(res, man.value, MPFR_RNDZ)

TypeError: Cannot convert gmpy2.mpz to sage.rings.integer.Integer

Hopefully it's only conversion. As I said, sage is expecting man to be a sage integer. Also when going in the other direction sage will use sage integers for the mantissa, maybe this breaks mpmath in other ways.

I see. Sage has additional cython replacements for mpmath functions, but this interface is outdated already for the 1.3.0 version (see e.g. sagemath/sage#36447). I think it might be dropped.

which version of scipy is compatible with this branch?

I am seeing

  File "/usr/lib/python3.11/site-packages/sympy/core/numbers.py", line 4567, in <module>
    _sympy_converter[type(mpmath.rational.mpq(1, 2))] = sympify_mpmath_mpq
                          ^^^^^^^^^^^^^^^
AttributeError: module 'mpmath' has no attribute 'rational'

while running ./sage -tp src/sage/libs/mpmath, so there is something going on there.

which version of scipy is compatible with this branch?

AFAIC, only the sympy master, since sympy/sympy#26269

Actually there is sympy 1.12.1a1 that should be compatible:
https://pypi.org/project/sympy/#history

Soon there should be 1.12.1

oh yeah, it's SymPy, not SciPy

Actually there is sympy 1.12.1a1 that should be compatible: https://pypi.org/project/sympy/#history

Yes, thanks for that. It's why I started testing this only now.

I don't know how easily this fits your workflow but you might find it easier to track mpmath and sympy master branches. SymPy's master branch tracks mpmath's master branch directly regardless of prereleases:
https://github.com/sympy/sympy/blob/17639a46b648f265c968ea86b78725258024f715/.github/workflows/runtests.yml#L461
It is more likely that the master branches are in sync than that prereleases are.

I don't know how easily this fits your workflow but you might find it easier to track mpmath and sympy master branches. SymPy's master branch tracks mpmath's master branch directly regardless of prereleases: https://github.com/sympy/sympy/blob/17639a46b648f265c968ea86b78725258024f715/.github/workflows/runtests.yml#L461 It is more likely that the master branches are in sync than that prereleases are.

I track prereleases to have ahead-of-time warning of what will happen with future updates (maiainly sagemath and dependencies, e.g. void-linux/void-packages#49571). Tracking master would be unworkable (and for most projects building from master is a different procedure than building from a release or prerelease, so it wouldn't be testing the same). I dogfood my own draft packages so they have to be in a usable combination.

I have the necessary update for Sage in sagemath/sage#38113