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

stillwater-sc / universal / 42118868265
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/fixpnt-quire-resolve-saturates
DEFAULT BRANCH: master
Ran 05 Nov 2026 10:19PM UTC
Jobs 1
Files 842
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

05 Nov 2026 10:08PM UTC coverage: 88.567% (-0.005%) from 88.572%
42118868265

push

github

web-flow
fix(dbns): convert to the nearest encoding, in every configuration (#1939)

dbns<16,5> converted 1.5 to 1.4e122 and dbns<16,8> to -0.25. Conversion
to and from double had several defects, which reached far beyond those two
configurations:
- The encoding was written into the most significant block alone, so every
  dbns wider than its block type converted almost every value wrongly. That
  includes dbns<10,4> and up with the default uint8_t: 1020 of 1022
  encodings failed a round trip.
- The (a, b) search admitted out-of-range first base exponents and
  projected them back badly, so many values landed on zero or a far
  neighbour even in single-block types. It now clamps to the fields, skips
  the zero pattern, keeps the pair nearest in log2, and searches only a
  bounded window in 64-bit arithmetic. dbns<34,1> had not searched at all.
- minpos() produced the zero pattern, so minpos was 0 and minneg NaN.
- setexponent() shifted in 32 bits, which lost fields in one uint64_t
  block.
- Decoding computed 0.5^a and 3^b apart, so an in-range value whose factors
  left the range came out inf, 0 or NaN. It now uses 2^L, with a split
  log2(3), and keeps long double's range and float subnormals at compile
  time.
- Static asserts bound both fields to 32 bits, and one field to 20, which
  keeps a conversion near a million steps at most.

New test static/logarithmic/dbns/conversion/nearest.cpp covers:
- exhaustive round trips;
- the brute-force nearest encoding;
- wide fields against an oracle;
- single- and multi-block layouts;
- minpos and underflow;
- constexpr decodes.
main fails 12 of its original 15 checks.

Resolves #1933

BEHAVIOUR-CHANGE: dbns conversion from double now returns the nearest encoding (multi-block configurations such as dbns<16,5> were wrong for almost every value), decoding no longer overflows or underflows for in-range values, and minpos/minneg are 0.5^(MAX_A-1) instead of 0/NaN
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

49 of 57 new or added lines in 1 file covered. (85.96%)

1 existing line in 1 file now uncovered.

57261 of 64653 relevant lines covered (88.57%)

8911281.93 hits per line

Uncovered Changes

Lines Coverage ∆ File
8
91.59
-1.35% include/sw/universal/number/dbns/dbns_impl.hpp

Coverage Regressions

Lines Coverage ∆ File
1
97.8
0.0% include/sw/universal/number/cfloat/cfloat_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 42118868265.1 05 Nov 2026 10:19PM UTC 842
88.57
GitHub Action Run
Source Files on build 42118868265
  • Tree
  • List 842
  • Changed 3
  • Source Changed 1
  • Coverage Changed 3
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #42118868265
  • 7bc791e6 on github
  • Prev Build on main (#42103547817)
  • Next Build on main (#42125924563)
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