moonbitlang / core / 6620

89%

push

github

math: return 1 for pow(1, NaN) and pow(±1, ±inf) per IEEE 754 (#4219)

* math: return 1 for pow(1, NaN) and pow(+-1, +-inf) per IEEE 754

IEEE 754-2019 section 9.2.1 / C99 Annex F.9.4.4 specify pow(1, y) = 1
for any y (even a NaN) and pow(-1, +-inf) = 1. The Double pow returned
NaN for these on every backend (fdlibm NaN check first; the js backend
delegates to ECMAScript Math.pow), diverging from its own doc comment
and from powf, which already implements the IEEE behaviour.

Align the non-js fdlibm port and wrap the js FFI with the unit-base
special cases. pow(-1, NaN) is unspecified by the standard and stays
NaN. The quickcheck special-value suite is flipped to the IEEE
behaviour as suggested in #4059.

* math: remove unreachable pow unit-base infinity branch

---------

Co-authored-by: Hongbo Zhang <bobzhang1988@gmail.com>

3 of 3 new or added lines in 1 file covered. (100.0%)

16441 of 18408 relevant lines covered (89.31%)

273252.69 hits per line