stillwater-sc / universal / 26397640216 / 1
84%
master: 84%

LAST BUILD BRANCH: fix/issue-ci-ereal-regression-runtime
DEFAULT BRANCH: master
Ran
Files 658
Run time 26s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 84.032% (+0.007%) from 84.025%
26397640216.1

push

github

web-flow 
fix(erational): correct ieee754 conversion (both directions) + add tests (#986) (#993)

* fix(erational): inbound double->erational conversion exponent scaling (#986, WIP)

convert_ieee754 now builds the exact dyadic rational
(-1)^s * significand * 2^(e-bias-fbits), handling normal, subnormal, and zero,
instead of dropping the exponent (which collapsed every value into [1,2)).
Verified at the rational level: erational(1)+erational(2)==erational(3) and
(1/3)*3==1 now hold.

WIP -- NOT a complete fix: the OUTBOUND conversion to_ieee754() is also broken
(ignores sign; naive edecimal->double of large numerator/denominator
overflows/loses precision), so double round-trips still fail. Needs a
correctly-rounded rational->double. Not for merge yet.

Relates to #986.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(erational): correct ieee754 conversion both directions + add tests (#986)

Inbound (double -> erational): convert_ieee754 now builds the exact dyadic
rational (-1)^s * significand * 2^(e-bias-fbits), handling normal, subnormal,
and zero. Previously it dropped the exponent entirely, collapsing every value
into [1,2) (erational(2.0)==1, erational(3.0)==1.5) and leaving subnormals/zero
unhandled (divide-by-zero).

Outbound (erational -> double): to_ieee754 now performs a correctly-rounded
(round-half-to-even) conversion of the big rational, with sign and subnormal
handling. The previous Real(numerator)/Real(denominator) ignored the sign and
overflowed/lost precision whenever numerator or denominator did not fit a
Real (subnormals, 0.1, 1e-300).

Add elastic/rational/decimal/conversion/ieee754.cpp: round-trip fuzz for double
and float, specific values (integers, fractions, signs, subnormals, DBL_MIN,
DBL_TRUE_MIN, DBL_MAX, 1e+/-300), and arithmetic round-trips. This is the
ieee754-conversion coverage erational never had -- the gap that let the bug
ship. (edecimal, which erational builds on, was probed sound; normalize() is a
... (continued)

47440 of 56455 relevant lines covered (84.03%)

8735897.35 hits per line

Source Files on job 26397640216.1