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

stillwater-sc / universal / 42161765292
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 06 Nov 2026 04:37AM UTC
Jobs 1
Files 852
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

06 Nov 2026 04:18AM UTC coverage: 88.605% (+0.002%) from 88.603%
42161765292

push

github

web-flow
feat(numeric): faithful's math library: classification, fmin/fmax and a compensated sqrt (#1953)

faithful's mathlib.hpp was another type's. It included 14 headers from a
faithful/math/ directory that did not exist, so it could not be
included, and the umbrella left it out. faithful had no math functions.

mathlib.hpp now includes three headers in math/functions/, and the
umbrella includes it:
- classify.hpp: isnan, isinf, isfinite, fpclassify, isnormal, isdenorm,
  iszero, signbit, copysign and fabs, of the represented value
  value + error. They read faithful's members, so they serve
  faithful<cfloat> too. signbit reports -0 as std::signbit does. A pair
  whose sum cancels, such as (1, -1) or (-1, 1), is +0.
- minmax.hpp: fmin and fmax, ordered exactly by faithful's comparisons. A
  NaN operand is ignored, as in std::fmin.
- sqrt.hpp: a compensated square root for a native FloatingPointType.
  The argument is scaled into [0.5, 2) by an exact even power of two.
  Real's sqrt follows, then one Newton correction from the residual
  b - s*s, exact through twoProd, then the result is scaled back, as in
  dd's sqrt. A negative argument gives EDOM and a quiet NaN, or a throw
  when FAITHFUL_THROW_ARITHMETIC_EXCEPTION is set.

numeric/faithful/mathlib.cpp checks the classification for float,
double and cfloat. It checks the sign functions and fmin/fmax, including
NaN, signed zeros and cancelling pairs, and the special values of sqrt.
It measures sqrt over 100000 arguments across the exponent range. The
reference is long double for faithful<float> and qd for faithful<double>.
The worst error is 1.32 and 1.37 units of epsilon(), against a bound of
2. Evaluating through Real gives about 10^7 and 10^15 units, and the
test fails it.

The other elementary functions need compensated algorithms of their own
and are #1952.

Resolves #1950

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

151 of 173 new or added lines in 4 files covered. (87.28%)

57893 of 65338 relevant lines covered (88.61%)

8849735.34 hits per line

Uncovered Changes

Lines Coverage ∆ File
22
82.95
numeric/faithful/mathlib.cpp
Jobs
ID Job ID Ran Files Coverage
1 42161765292.1 06 Nov 2026 04:37AM UTC 852
88.61
GitHub Action Run
Source Files on build 42161765292
  • Tree
  • List 852
  • Changed 3
  • Source Changed 0
  • Coverage Changed 3
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #42161765292
  • dc38d2e7 on github
  • Prev Build on main (#42155020823)
  • Next Build on main (#42166969421)
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