Skip to content
Docs Pydantic Docs

decimal module

Monty’s decimal is a port of CPython’s pure-Python _pydecimal with the behaviour of the C _decimal module CPython actually ships: arithmetic results, the flags each operation sets, which signal a trapped operation raises and its [<class 'decimal.DivisionByZero'>] message all follow _decimal. Decimal, Context, localcontext(), getcontext() / setcontext(), the three template contexts, IEEEContext(), DecimalTuple, every signal class and every constant are available.

Contexts

  • There is one current context per session, not per thread or per contextvars context. localcontext() and setcontext() work as in CPython, and the current context persists across REPL feeds and snapshots like global variables do.
  • ctx.flags and ctx.traps return a new mapping object on every access, so ctx.flags is ctx.flags is False; the mappings read and write through to the context, so everything else behaves the same. Their type prints as <class 'SignalDict'> (CPython: <class 'abc.SignalDict'>) and is not a collections.abc.MutableMapping. keys(), values() and items() return lists rather than views. A mapping keeps its context alive, so it stays usable after the context is gone, where CPython 3.14.9+ raises ValueError: invalid signal dict (as in copy.deepcopy(ctx).flags[Inexact]).
  • Context objects cannot be pickled (__reduce__ is absent); copy.copy() and copy.deepcopy() work.

Decimal

  • Decimal and Context cannot be subclassed (Monty has no class inheritance).
  • A coefficient (or NaN payload) longer than 4300 digits is rejected when a Decimal is built from a str, int or tuple, with ValueError: Exceeds the limit (4300 digits) for a Decimal coefficient: value has N digits. The same applies to an int operand in arithmetic (Decimal(1) + 10**5000). CPython has no such limit. Results computed under a large prec are not capped, but materializing one past the memory limit raises MemoryError.
  • hash() values differ from CPython’s, as Monty’s int and float hashes do; equal numbers still hash alike (hash(Decimal('1.5')) == hash(1.5), hash(Decimal(2)) == hash(2)), except that an integral Decimal with more than 5000 digits does not hash like the equal int.
  • A trapped signal’s exception carries its message as a string: exc.args is ("[<class 'decimal.DivisionByZero'>]",) where CPython’s is a one-element tuple holding a list of classes. str(exc) matches.
  • Decimal.from_float and Decimal.from_number can only be called in place (Decimal.from_float(0.1)); reading them as attributes (getattr(Decimal, 'from_float'), f = Decimal.from_float) raises AttributeError, as for Monty’s other builtin class methods.
  • Decimal.as_tuple() returns a plain tuple subclass whose type is not decimal.DecimalTuple (isinstance(d.as_tuple(), DecimalTuple) is False); the sign, digits and exponent fields work.

Formatting

  • Decimal.__format__(spec, override) checks that override is a dict but ignores its thousands_sep, decimal_point and grouping keys.

json

json.loads(s, parse_float=Decimal) works (see json.md); json.dumps rejects a Decimal with TypeError: Object of type Decimal is not JSON serializable, as CPython does.