fix mpmath precision handling (#781)

This PR fixes another issue with precision. In particular, this issue
was the reason of the hangings in doctest in #778.

mathics/builtin/arithmetic.py

@@ -6,6 +6,7 @@
 
 Basic arithmetic functions, including complex number arithmetic.
 """
+from typing import Callable, Optional
 
 from mathics.eval.numerify import numerify
 
@@ -52,7 +53,12 @@
 from mathics.core.element import ElementsProperties
 from mathics.core.expression import Expression
 from mathics.core.list import ListExpression
-from mathics.core.number import SpecialValueError, dps, min_prec
+from mathics.core.number import (
+    FP_MANTISA_BINARY_DIGITS,
+    SpecialValueError,
+    dps,
+    min_prec,
+)
 from mathics.core.symbols import (
     Atom,
     Symbol,
@@ -68,9 +74,7 @@
     SymbolAnd,
     SymbolComplexInfinity,
     SymbolDirectedInfinity,
-    SymbolIndeterminate,
     SymbolInfix,
-    SymbolOverflow,
     SymbolPiecewise,
     SymbolPossibleZeroQ,
     SymbolTable,
@@ -79,20 +83,35 @@
 from mathics.eval.nevaluator import eval_N
 
 
-@lru_cache(maxsize=4096)
-def call_mpmath(mpmath_function, mpmath_args):
-    try:
-        return mpmath_function(*mpmath_args)
-    except ValueError as exc:
-        text = str(exc)
-        if text == "gamma function pole":
-            return SymbolComplexInfinity
-        else:
-            raise
-    except ZeroDivisionError:
-        return
-    except SpecialValueError as exc:
-        return Symbol(exc.name)
+# @lru_cache(maxsize=4096)
+def call_mpmath(
+    mpmath_function: Callable, mpmath_args: tuple, prec: Optional[int] = None
+):
+    """
+    calls the mpmath_function with mpmath_args parms
+    if prec=None, use floating point arithmetic.
+    Otherwise, work with prec bits of precision.
+    """
+    # TODO: rocky, please help me with the annotations
+    # in the signature of this function.
+    if prec is None:
+        prec = FP_MANTISA_BINARY_DIGITS
+    with mpmath.workprec(prec):
+        try:
+            result_mp = mpmath_function(*mpmath_args)
+            if prec != FP_MANTISA_BINARY_DIGITS:
+                return from_mpmath(result_mp, prec)
+            return from_mpmath(result_mp)
+        except ValueError as exc:
+            text = str(exc)
+            if text == "gamma function pole":
+                return SymbolComplexInfinity
+            else:
+                raise
+        except ZeroDivisionError:
+            return
+        except SpecialValueError as exc:
+            return Symbol(exc.name)
 
 
 class _MPMathFunction(SympyFunction):
@@ -147,38 +166,18 @@ def eval(self, z, evaluation):
                 return
 
             result = call_mpmath(mpmath_function, tuple(float_args))
-
-            if isinstance(result, (mpmath.mpc, mpmath.mpf)):
-                if mpmath.isinf(result) and isinstance(result, mpmath.mpc):
-                    result = SymbolComplexInfinity
-                elif mpmath.isinf(result) and result > 0:
-                    result = Expression(SymbolDirectedInfinity, Integer1)
-                elif mpmath.isinf(result) and result < 0:
-                    result = Expression(SymbolDirectedInfinity, IntegerM1)
-                elif mpmath.isnan(result):
-                    result = SymbolIndeterminate
-                else:
-                    # FIXME: replace try/except as a context manager
-                    # like "with evaluation.from_mpmath()...
-                    # which can be instrumented for
-                    # or mpmath tracing and benchmarking on demand.
-                    # Then use it on other places where mpmath appears.
-                    try:
-                        result = from_mpmath(result)
-                    except OverflowError:
-                        evaluation.message("General", "ovfl")
-                        result = Expression(SymbolOverflow)
         else:
             prec = min_prec(*args)
             d = dps(prec)
             args = [eval_N(arg, evaluation, Integer(d)) for arg in args]
+
             with mpmath.workprec(prec):
+                # to_mpmath seems to require that the precision is set from outside
                 mpmath_args = [x.to_mpmath() for x in args]
                 if None in mpmath_args:
                     return
-                result = call_mpmath(mpmath_function, tuple(mpmath_args))
-                if isinstance(result, (mpmath.mpc, mpmath.mpf)):
-                    result = from_mpmath(result, precision=prec)
+
+                result = call_mpmath(mpmath_function, tuple(mpmath_args), prec)
         return result
 
 
@@ -723,7 +722,7 @@ def to_sympy(self, expr, **kwargs):
             return sympy.zoo
 
 
-class I(Predefined):
+class I_(Predefined):
     """
     <url>:Imaginary unit:https://en.wikipedia.org/wiki/Imaginary_unit</url> \
     (<url>:WMA:https://reference.wolfram.com/language/ref/I.html</url>)
@@ -739,6 +738,7 @@ class I(Predefined):
      = 10
     """
 
+    name = "I"
     summary_text = "imaginary unit"
     python_equivalent = 1j
 

mathics/builtin/numbers/constants.py

@@ -27,6 +27,7 @@
     MIN_MACHINE_NUMBER,
     PrecisionValueError,
     get_precision,
+    prec,
 )
 from mathics.core.symbols import Atom, Symbol, strip_context
 from mathics.core.systemsymbols import SymbolIndeterminate
@@ -43,8 +44,11 @@ def mp_constant(fn: str, d=None) -> mpmath.mpf:
         # ask for a certain number of digits, but the
         # accuracy will be less than that. Figure out
         # what's up and compensate somehow.
-        mpmath.mp.dps = int_d = int(d * 3.321928)
-        return getattr(mpmath, fn)(prec=int_d)
+
+        int_d = prec(d)
+        with mpmath.workprec(int_d):
+            result = str(getattr(mpmath, fn)(prec=int_d))
+            return result
 
 
 def mp_convert_constant(obj, **kwargs):

mathics/builtin/specialfns/gamma.py

@@ -112,17 +112,6 @@ def eval_with_z(self, z, a, b, evaluation):
                 return
 
             result = call_mpmath(mpmath_function, tuple(float_args))
-            if isinstance(result, (mpmath.mpc, mpmath.mpf)):
-                if mpmath.isinf(result) and isinstance(result, mpmath.mpc):
-                    result = SymbolComplexInfinity
-                elif mpmath.isinf(result) and result > 0:
-                    result = Expression(SymbolDirectedInfinity, Integer1)
-                elif mpmath.isinf(result) and result < 0:
-                    result = Expression(SymbolDirectedInfinity, Integer(-1))
-                elif mpmath.isnan(result):
-                    result = SymbolIndeterminate
-                else:
-                    result = from_mpmath(result)
         else:
             prec = min_prec(*args)
             d = dps(prec)
@@ -131,9 +120,7 @@ def eval_with_z(self, z, a, b, evaluation):
                 mpmath_args = [x.to_mpmath() for x in args]
                 if None in mpmath_args:
                     return
-                result = call_mpmath(mpmath_function, tuple(mpmath_args))
-                if isinstance(result, (mpmath.mpc, mpmath.mpf)):
-                    result = from_mpmath(result, precision=prec)
+                result = call_mpmath(mpmath_function, tuple(mpmath_args), prec)
         return result
 
 

mathics/core/atoms.py

@@ -249,13 +249,18 @@ def to_sympy(self, **kwargs):
     def to_python(self, *args, **kwargs):
         return self.value
 
-    def round(self, d=None) -> Union["MachineReal", "PrecisionReal"]:
+    def round(self, d: Optional[int] = None) -> Union["MachineReal", "PrecisionReal"]:
+        """
+        Produce a Real approximation of ``self`` with decimal precision ``d``.
+        If ``d`` is  ``None``, and self.value fits in a float,
+        returns a ``MachineReal`` number.
+        Is the low-level equivalent to ``N[self, d]``.
+        """
         if d is None:
             d = self.value.bit_length()
             if d <= FP_MANTISA_BINARY_DIGITS:
                 return MachineReal(float(self.value))
             else:
-                # FP_MANTISA_BINARY_DIGITS / log_2(10) + 1
                 d = MACHINE_PRECISION_VALUE
         return PrecisionReal(sympy.Float(self.value, d))
 
@@ -441,7 +446,10 @@ def make_boxes(self, form):
     def is_zero(self) -> bool:
         return self.value == 0.0
 
-    def round(self, d=None) -> "MachineReal":
+    def round(self, d: Optional[int] = None) -> "MachineReal":
+        """
+        Produce a Real approximation of ``self`` with decimal precision ``d``.
+        """
         return self
 
     def sameQ(self, other) -> bool:
@@ -540,12 +548,11 @@ def make_boxes(self, form):
             self, dps(self.get_precision()), None, None, _number_form_options
         )
 
-    def round(self, d=None) -> Union[MachineReal, "PrecisionReal"]:
+    def round(self, d: Optional[int] = None) -> Union[MachineReal, "PrecisionReal"]:
         if d is None:
             return MachineReal(float(self.value))
-        else:
-            d = min(dps(self.get_precision()), d)
-            return PrecisionReal(self.value.n(d))
+        _prec = min(prec(d), self.value._prec)
+        return PrecisionReal(sympy.Float(self.value, precision=_prec))
 
     def sameQ(self, other) -> bool:
         """Mathics SameQ for PrecisionReal"""

mathics/core/convert/mpmath.py

@@ -6,8 +6,17 @@
 import mpmath
 import sympy
 
-from mathics.core.atoms import Complex, MachineReal, MachineReal0, PrecisionReal
+from mathics.core.atoms import (
+    Complex,
+    Integer1,
+    IntegerM1,
+    MachineReal,
+    MachineReal0,
+    PrecisionReal,
+)
+from mathics.core.expression import Expression
 from mathics.core.symbols import Atom
+from mathics.core.systemsymbols import SymbolDirectedInfinity
 
 
 @lru_cache(maxsize=1024)

mathics/eval/nevaluator.py

@@ -13,13 +13,13 @@
 
 import sympy
 
-from mathics.core.atoms import Number
+from mathics.core.atoms import Number, PrecisionReal
 from mathics.core.attributes import A_N_HOLD_ALL, A_N_HOLD_FIRST, A_N_HOLD_REST
 from mathics.core.convert.sympy import from_sympy
 from mathics.core.element import BaseElement
 from mathics.core.evaluation import Evaluation
 from mathics.core.expression import Expression
-from mathics.core.number import PrecisionValueError, get_precision
+from mathics.core.number import PrecisionValueError, dps, get_precision
 from mathics.core.symbols import Atom
 from mathics.core.systemsymbols import SymbolMachinePrecision, SymbolN
 
@@ -50,7 +50,6 @@ def eval_NValues(
     stored in ``evaluation.definitions``.
     If ``prec`` can not be evaluated as a number, returns None, otherwise, returns an expression.
     """
-
     # The first step is to determine the precision goal
     try:
         # Here ``get_precision`` is called with ``show_messages``

test/builtin/atomic/test_numbers.py

@@ -270,3 +270,38 @@ def test_accuracy(str_expr, str_expected):
 )
 def test_precision(str_expr, str_expected):
     check_evaluation(f"Precision[{str_expr}]", str_expected)
+
+
+@pytest.mark.parametrize(
+    ("str_expr", "str_expected", "msg"),
+    [
+        (None, None, None),
+        ("N[Sqrt[2], 41]//Precision", "41.", "first round sqrt[2`41]"),
+        ("N[Sqrt[2], 40]//Precision", "40.", "first round sqrt[2`40]"),
+        ("N[Sqrt[2], 41]//Precision", "41.", "second round sqrt[2`41]"),
+        ("N[Sqrt[2], 40]//Precision", "40.", "second round sqrt[2`40]"),
+        (
+            "N[Sqrt[2], 41]",
+            '"1.4142135623730950488016887242096980785697"',
+            "third round sqrt[2`41]",
+        ),
+        (
+            "Precision/@Table[N[Pi,p],{p, {5, 100, MachinePrecision, 20}}]",
+            "{5., 100., MachinePrecision, 20.}",
+            None,
+        ),
+        (
+            "Precision/@Table[N[Sin[1],p],{p, {5, 100, MachinePrecision, 20}}]",
+            "{5., 100., MachinePrecision, 20.}",
+            None,
+        ),
+        ("N[Sqrt[2], 40]", '"1.414213562373095048801688724209698078570"', None),
+        ("N[Sqrt[2], 4]", '"1.414"', None),
+        ("N[Pi, 40]", '"3.141592653589793238462643383279502884197"', None),
+        ("N[Pi, 4]", '"3.142"', None),
+        ("N[Pi, 41]", '"3.1415926535897932384626433832795028841972"', None),
+        ("N[Sqrt[2], 41]", '"1.4142135623730950488016887242096980785697"', None),
+    ],
+)
+def test_change_prec(str_expr, str_expected, msg):
+    check_evaluation(str_expr, str_expected, failure_message=msg)