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

stillwater-sc / universal / 42337764573
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 07 Nov 2026 04:33AM 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

07 Nov 2026 04:17AM UTC coverage: 88.662% (+0.02%) from 88.64%
42337764573

push

github

web-flow
fix(interval): enclose subnormal square roots and handle unbounded and empty intervals (#1979)

* fix(interval): enclose subnormal square roots and handle unbounded and empty intervals

The QA sweep (#1962) found interval losing containment or returning NaN
at the edges of its domain. Each case reproduced on main.

- sqrt: the residual s*s - x decides which endpoint to widen, and fma
  gives it exactly only while it is representable. For a small x its
  bits fall below denorm_min, it rounds to zero, and an inexact root was
  taken as exact: 1955 of the 1999 values k*denorm_min excluded their
  root. The new sqrt_enclose forms the residual after an exact scaling
  by 2^(2 digits) below min() * 2^(2 digits). For cfloat it decides by
  exact double promotion, as prod_enclose does. That also fixes
  interval<cfloat<16,5>> with subnormals, whose fma residual was not
  exact at 11 bits: 1939 of its 31743 positive encodings lost
  containment.
- sqrt of a fully negative interval is the empty interval (NaN bounds)
  with EDOM. It was [0, NaN].
- mid(): two large bounds of one sign halve before adding. mid of
  [1e308, 1.7e308] was inf. An unbounded interval follows IEEE
  1788-2015: [-inf, +inf] gives 0 (it gave NaN, and so did double()),
  and a half-bounded one gives the finite extreme on its open side.
- The empty interval is unordered: <= and >= were both true for it.
  hull(empty, x) is x in either order (it was [nan, nan] in one), and
  intersect(x, empty) is empty (it was x).
- pow with an odd n is [lo^n, hi^n]. x * x^(n-1) treated the factors as
  independent: pow([-2, 3], 3) was [-18, 27] for the exact [-8, 27].
- operator>> reads [lo, hi] or sets failbit and leaves its target
  unchanged. It overwrote the target on a failed parse and accepted any
  delimiters, such as "[1, 2)".

New test static/range/interval/arithmetic/special_values.cpp:
- subnormal sqrt for float and double, with containment decided
  exactly by an error-free square in a scaled d... (continued)

96 of 97 new or added lines in 2 files covered. (98.97%)

59267 of 66846 relevant lines covered (88.66%)

9628357.76 hits per line

Uncovered Changes

Lines Coverage ∆ File
1
98.18
-1.82% include/sw/universal/number/interval/iostream.hpp
Jobs
ID Job ID Ran Files Coverage
1 42337764573.1 07 Nov 2026 04:33AM UTC 868
88.66
GitHub Action Run
Source Files on build 42337764573
  • Tree
  • List 868
  • Changed 3
  • Source Changed 2
  • Coverage Changed 3
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #42337764573
  • 608d5601 on github
  • Prev Build on main (#42315031681)
  • Next Build on main (#42337781619)
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