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

stillwater-sc / universal / 42297712065
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 06 Nov 2026 09:53PM UTC
Jobs 1
Files 868
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 09:33PM UTC coverage: 88.636% (+0.007%) from 88.629%
42297712065

push

github

web-flow
fix(numeric): faithful divides by infinity, keeps -0, and pow underflows (#1974)

The QA sweep (#1962) found that a finite faithful divided by an infinite
one is NaN, and that pow with a negative integer exponent is NaN
whenever the positive power overflows.

- operator/=: with a non-finite head on either side, the result is the
  IEEE quotient of the heads with no error. twoDiv's residual
  a - b*d is 1 - inf*0 = NaN for 1/inf, and the error term turned the
  finite quotient 0 into NaN. Division is the only operator that takes
  an infinite operand to a finite head; + - * give a non-finite head
  there, whose error renormalize() already clears.
- renormalize(): a pair with a zero error is canonical and is left
  alone. twoSum(-0, 0) is +0, so a -0 head lost its sign: -0 * 1 was +0,
  and x / -0 divided by +0 and gave the wrong infinity. The new test
  found this.
- ipow: x = 2^k b with b in [sqrt(1/2), sqrt(2)) is raised as b^|n|,
  which stays within 2^(+-32), and 2^(k n) is applied once at the end,
  with the head's remainder joining the error part where the result is
  subnormal. pow(2^20, -52) was 1 / inf; it is now 2^-1040. In range
  the result is unchanged: the pow accuracy figures are identical to
  main's in every row.

numeric/faithful/compensated_arithmetic.cpp checks + - * / and /= for
every pair of finite, zero, infinite and NaN operands against IEEE
arithmetic on the heads, division by +-0, and signed-zero results, for
float and double. hyperbolic_and_power.cpp checks a negative exponent
whose positive power overflows, with results at 0 (ERANGE) and at
subnormal values. Reverting each fix fails its test: 28, 15 and 2 cases
per type. The subnormal pow result is held to one subnormal ulp
against a dd reference (CodeRabbit).

The issue's remaining item, measuring the per-call allocations of
exact_expansion, is not addressed here.

Relates to #1969
Relates to #1962

BEHAVIOUR-CHANGE: faithful finite / +-inf is now a signed zero instead of N... (continued)

72 of 75 new or added lines in 4 files covered. (96.0%)

1 existing line in 1 file now uncovered.

59093 of 66669 relevant lines covered (88.64%)

9641974.55 hits per line

Uncovered Changes

Lines Coverage ∆ File
3
80.63
3.77% numeric/faithful/compensated_arithmetic.cpp

Coverage Regressions

Lines Coverage ∆ File
1
97.8
-0.06% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 42297712065.1 06 Nov 2026 09:53PM UTC 868
88.64
GitHub Action Run
Source Files on build 42297712065
  • Tree
  • List 868
  • Changed 6
  • Source Changed 4
  • Coverage Changed 6
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #42297712065
  • f3c510fe on github
  • Prev Build on main (#42291239709)
  • Next Build on main (#42312921512)
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