• Home
  • Features
  • Pricing
  • Docs
  • Announcements
  • Sign In

stillwater-sc / universal / 42523560319
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 08 Nov 2026 09:28PM UTC
Jobs 1
Files 871
Run time 1min
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

08 Nov 2026 09:07PM UTC coverage: 88.835% (+0.03%) from 88.802%
42523560319

push

github

web-flow
fix(ereal): C99 special values, exact fmod, and math tests checked to the type's precision (#2018)

A probe of every ereal math function against libm found 427 mismatches
at NaN, +-inf, +-0 and the domain edges. Among them:
- sqrt(-1) was -1, and sqrt(+inf) and log(+inf) were NaN.
- hypot(1e200, 1e200) was NaN.
- pow(0, -1), pow(-2, 1e200) and atan2(0, -0) were NaN.
- fmod(5, inf) was NaN.
- sin(-0), floor(-0), rint(-0.3) and modf(-3) lost the sign of zero.
- copysign read -0 as positive, and nextafter(NaN, y) returned y.
Every function now returns the C99 Annex F values. The shared helpers
are in math/functions/special_values.hpp.

Other fixes:
- fmod and remainder are exact for quotients of any size, with a
  chunked, division-free reduction. fmod(1e200, 3) was 1.4e70.
- fmod(x, 0) and remainder(x, 0) return NaN (user decision). They throw
  only under EREAL_THROW_ARITHMETIC_EXCEPTION, as x / 0 does.
- hypot scales its operands by a power of two first.
- asinh and acosh use an overflow-free form above 2^(max_exponent/4).
  The fixed 2^256 limit let float limbs overflow.
- asin switches to its sqrt identity at 1/sqrt(2) instead of 0.8. The
  series hit its iteration cap and lost up to 100 bits for
  0.7 < |x| <= 0.8.
- sqrt uses Karp's division-free method: 1.9 -> 0.5 ms at ereal<8>, for
  the same digits. Perfect squares are now exact.

The math tests compared through double with tolerances of 1e-10 to
1e-15, which an ereal<8> result correct to 53 bits passes (#2008).
They now check agreement in bits, exactly, with the new
verification/ereal_precision.hpp. References are stored constants, an
ereal<19> result, mpmath values or identities. A value truncated to 53
bits fails these checks. The conversion found erf, erfc, tgamma, lgamma
and complex abs/arg accurate only to double (#2017). Large-argument
trigonometry is filed as #2016.

New test: elastic/ereal/math/functions/c99_special_values.cpp.

BEHAVIOUR-CHANGE: ereal math functions return C99 Annex ... (continued)

216 of 227 new or added lines in 14 files covered. (95.15%)

4 existing lines in 2 files now uncovered.

59959 of 67495 relevant lines covered (88.83%)

9707120.71 hits per line

Uncovered Changes

Lines Coverage ∆ File
9
76.32
include/sw/universal/verification/ereal_precision.hpp
1
98.04
8.57% include/sw/universal/number/ereal/math/functions/fractional.hpp
1
95.24
include/sw/universal/number/ereal/math/functions/special_values.hpp

Coverage Regressions

Lines Coverage ∆ File
2
84.93
-0.3% include/sw/universal/number/qd/qd_impl.hpp
2
76.74
-4.65% include/sw/universal/verification/test_suite_mathlib_adaptive.hpp
Jobs
ID Job ID Ran Files Coverage
1 42523560319.1 08 Nov 2026 09:28PM UTC 871
88.83
GitHub Action Run
Source Files on build 42523560319
  • Tree
  • List 871
  • Changed 19
  • Source Changed 13
  • Coverage Changed 18
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #42523560319
  • 25778d2f on github
  • Prev Build on main (#42504241940)
  • Next Build on main (#42529495092)
STATUS · Troubleshooting · Open an Issue · Sales · Support · CAREERS · ENTERPRISE · START FREE TRIAL · SCHEDULE DEMO
ANNOUNCEMENTS · TWITTER · TOS & SLA · Supported CI Services · What's a CI service? · Automated Testing

© 2026 Coveralls, Inc