pynac bug -- evaluation of arctan2 fails with traceback

[williamstein]

~/Sara$ sage-develop
┌────────────────────────────────────────────────────────────────────┐
│ SageMath version 7.6.beta5, Release Date: 2017-02-26               │
│ Type "notebook()" for the browser-based notebook interface.        │
│ Type "help()" for help.                                            │
└────────────────────────────────────────────────────────────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Warning: this is a prerelease version, and it may be unstable.     ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
sage: arctan2(1.5,-1.300000000000001)
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
<ipython-input-1-ccfa336aad7e> in <module>()
----> 1 arctan2(RealNumber('1.5'),-RealNumber('1.300000000000001'))
 
/projects/sage/sage-dev/src/sage/symbolic/function.pyx in sage.symbolic.function.BuiltinFunction.__call__ (/projects/sage/sage-dev/src/build/cythonized/sage/symbolic/function.cp
p:11445)()
    994             res = self._evalf_try_(*args)
    995             if res is None:
--> 996                 res = super(BuiltinFunction, self).__call__(
    997                         *args, coerce=coerce, hold=hold)
    998
 
/projects/sage/sage-dev/src/sage/symbolic/function.pyx in sage.symbolic.function.Function.__call__ (/projects/sage/sage-dev/src/build/cythonized/sage/symbolic/function.cpp:6589)
()
    489                     (<Expression>args[0])._gobj, hold)
    490         elif self._nargs == 2:
--> 491             res = g_function_eval2(self._serial, (<Expression>args[0])._gobj,
    492                     (<Expression>args[1])._gobj, hold)
    493         elif self._nargs == 3:
 
/projects/sage/sage-dev/src/sage/libs/pynac/pynac.pyx in sage.libs.pynac.pynac.py_atan2 (/projects/sage/sage-dev/src/build/cythonized/sage/libs/pynac/pynac.cpp:21075)()
   1790     if P is float and parent(y) is not float:
   1791         P = RR
-> 1792     assert P is parent(y)
   1793     if P is ZZ:
   1794         P = RR
 
AssertionError:

In fact, pretty much any high precision input (even just padding with zeros) breaks this:

sage: atan2(2.1000000000000000000000000000000000000,-1.20000000000000000000000000000000)
---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)

Reported by: Sara Billey

CC: @paulmasson @rwst

Component: calculus

Author: Ralf Stephan, Jeroen Demeyer

Branch/Commit: a41922c

Reviewer: Jeroen Demeyer, Paul Masson

Issue created by migration from https://trac.sagemath.org/ticket/22553

[rwst]

comment:3

This was introduced in #21371. Not Pynac but its Python interface to Sage's numeric functions.

[rwst]

Branch: u/rws/pynac_bug____evaluation_of_arctan2_fails_with_traceback

[rwst]

Author: Ralf Stephan

[rwst]

Commit: 64e05e5

[rwst]

New commits:

64e05e5

22553: remove assert

[jdemeyer]

comment:6

Why not use

from sage.structure.element cimport coercion_model

P = coercion_model.common_parent(x, y)

[jdemeyer]

Reviewer: Jeroen Demeyer

[jdemeyer]

comment:7

I'll make that change and some further changes.

[jdemeyer]

Changed branch from u/rws/pynac_bug____evaluation_of_arctan2_fails_with_traceback to u/jdemeyer/pynac_bug____evaluation_of_arctan2_fails_with_traceback

[jdemeyer]

Changed commit from 64e05e5 to a41922c

[jdemeyer]

New commits:

a41922c

Minor fixes to py_atan2

[paulmasson]

comment:10

Jeroen, nice work on this. I've been devising ways to get around this assertion error for some time so it's nice to be rid of it. This is an elegant solution and the code cleanup looks good.

A run of make ptestlong produced the following errors:

sage -t --long src/sage/libs/ppl.pyx  # Timed out after testing finished
sage -t --long src/sage/misc/sagedoc.py  # 3 doctests failed

The ppl.pyx error went away on a subsequent test of just that file. Strangely enough, so did the sagedoc.py errors, so I guess we're good to go.

[paulmasson]

Changed author from Ralf Stephan to Ralf Stephan, Jeroen Demeyer

[paulmasson]

Changed reviewer from Jeroen Demeyer to Jeroen Demeyer, Paul Masson

[vbraun]

Changed branch from u/jdemeyer/pynac_bug____evaluation_of_arctan2_fails_with_traceback to a41922c