Lib/_pydecimal.py / Modules/_decimal/ (part 2)

Source:

cpython 3.14 @ ab2d84fe1023/Modules/_decimal/libmpdec/mpdecimal.c ↗

This annotation covers arithmetic operations and context management. See lib_decimal_detail for Decimal.__new__, __str__, __repr__, from_float, and the digit storage layout.

Map

Lines	Symbol	Role
1-120	Decimal.__add__ / __sub__	Aligned addition and subtraction
121-280	Decimal.__mul__	Schoolbook and Karatsuba multiplication
281-440	Decimal.__truediv__ / __floordiv__	Long division with rounding
441-620	Decimal.__pow__	Exponentiation with arbitrary precision
621-800	Context	precision, rounding, Emax, Emin, clamp, traps
801-1000	localcontext	Thread-local context manager
1001-1400	Signals and traps	Inexact, Overflow, DivisionByZero, InvalidOperation

Reading

Decimal.__add__

// CPython: Modules/_decimal/libmpdec/mpdecimal.c:4820 mpd_qadd
void
mpd_qadd(mpd_t *result, const mpd_t *a, const mpd_t *b,
         const mpd_context_t *ctx, uint32_t *status)
{
    /* Align a and b to the same exponent by shifting the smaller one.
       Then add coefficient arrays digit by digit. */
    mpd_t aligned;
    if (mpd_adjexp(a) < mpd_adjexp(b)) {
        _mpd_align(a, b, &aligned, ctx, status);
        _mpd_qbaseadd(result, &aligned, b, ctx, status);
    } else {
        _mpd_align(b, a, &aligned, ctx, status);
        _mpd_qbaseadd(result, a, &aligned, ctx, status);
    }
    mpd_qfinalize(result, ctx, status);  /* apply rounding */
}

Decimal addition aligns operands by adjusting their exponents, adds coefficient arrays, then rounds to the current context precision.

Context

# CPython: Lib/_pydecimal.py:380 Context
class Context:
    """Decimal arithmetic context.

    Attributes:
      prec        -- precision (default 28)
      rounding    -- rounding mode (default ROUND_HALF_EVEN)
      Emax        -- maximum exponent (default 999999999)
      Emin        -- minimum exponent (default -999999999)
      capitals    -- 1 = use E notation, 0 = use e
      clamp       -- 0 = do not clamp
      traps       -- dict mapping signal -> bool (raise vs. set flag)
      flags       -- dict mapping signal -> bool (accumulated signals)
    """

getcontext() returns the thread-local context. setcontext(ctx) installs a new one. decimal.ROUND_HALF_EVEN (banker's rounding) is the default.

localcontext

# CPython: Lib/_pydecimal.py:580 localcontext
class localcontext:
    """Context manager for a temporary decimal context."""
    def __init__(self, ctx=None, **kwargs):
        if ctx is None:
            self.saved_context = getcontext()
        else:
            self.saved_context = ctx
        self.new_context = self.saved_context.copy()
        for k, v in kwargs.items():
            setattr(self.new_context, k, v)

    def __enter__(self):
        setcontext(self.new_context)
        return self.new_context

    def __exit__(self, *args):
        setcontext(self.saved_context)

with decimal.localcontext() as ctx:
    ctx.prec = 50
    result = Decimal(1) / Decimal(3)

The context is restored even if an exception is raised.

Signals and traps

# CPython: Lib/_pydecimal.py:120 signals
class DecimalException(ArithmeticError): ...
class Clamped(DecimalException): ...
class InvalidOperation(DecimalException): ...
class DivisionByZero(DecimalException, ZeroDivisionError): ...
class Inexact(DecimalException): ...
class Overflow(Inexact, OverflowError): ...
class Underflow(Inexact): ...
class Subnormal(DecimalException): ...
class FloatOperation(DecimalException, TypeError): ...

Each signal has both a flag (set when the signal occurs) and a trap (raise an exception vs. record silently). ctx.traps[Overflow] = True causes overflow to raise Overflow immediately.

gopy notes

decimal uses libmpdec via Modules/_decimal. gopy's implementation is in module/decimal/module.go wrapping Go's github.com/shopspring/decimal or a custom mpd-based implementation. localcontext stores the context in contextvars.ContextVar. Signals map to gopy error codes in module/decimal/signals.go.