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

stillwater-sc / universal / 43767352162
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: fix/ci-pin-mingw-coverage-ubuntu-24.04
DEFAULT BRANCH: master
Ran 15 Nov 2026 10:25PM UTC
Jobs 1
Files 894
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

15 Nov 2026 10:05PM UTC coverage: 89.152% (+0.01%) from 89.14%
43767352162

push

github

web-flow
fix(cfloat): exact integer casts, round-once to_native, narrow and char sources, conversion across block types (#2090)

* fix(cfloat): exact integer casts, round-once to_native, narrow and char sources, conversion across block types

The cfloat core findings of sweep #2051:
- A negative short or signed char was negated in promoted int and
  sign-extended: cfloat<32,8,uint32_t,true,false,false>(short(-3)) was
  -9.22e18, with an undefined shift. The magnitude is now taken in the
  source's own width.
- Plain char took the unsigned path, so a signed char -1 became inf. It now
  converts as what it is on the platform.
- isminnegencoding() skipped a block from five blocks on: ++ on
  cfloat<40,8,uint8_t> 1.00000001.0..01 collapsed to +0. Its one-block case
  tested either bit instead of both.
- Conversion between block types went through double (maxpos -> inf,
  minpos -> 0 for cfloat<32,15>, and a static_assert past 52 fraction bits).
  The full-precision path reads the source independently of its block type,
  so it now serves every pair.
- to_int / to_long / to_long_long cast a double, losing every bit past 53.
  They now follow posit's rule (#2073), as takum and bposit do: truncate
  toward zero, exactly; saturate past the range, inf included; NaN -> the
  type's minimum (it was 0).
- to_native summed the fraction bit by bit in the target type and rounded
  twice: cfloat<40,8,uint32_t> 1.5 + 2^-24 + 2^-27 gave float 0x1.8p+0, not
  0x1.800002p+0, and double did the same past 52 fraction bits. It now
  rounds once, to nearest even, subnormal results included.
- fraction_ull() shifted a 64-bit limb by 64 (UB) on cfloat<80,16,uint64_t>
  and cfloat<72,11,uint32_t>. It reads only the blocks that start below fbits.
- shiftLeft() fell through to the block shift after zeroing for a shift past
  nbits.

Performance: to_native builds its result from limb-gathered windows of the
significand. cfloat -> double costs 10.4 ns instead of 128 for
cfloat<64,11,uint32_... (continued)

102 of 104 new or added lines in 1 file covered. (98.08%)

3 existing lines in 1 file now uncovered.

63496 of 71222 relevant lines covered (89.15%)

11936151.63 hits per line

Uncovered Changes

Lines Coverage ∆ File
2
98.3
0.44% include/sw/universal/number/cfloat/cfloat_impl.hpp

Coverage Regressions

Lines Coverage ∆ File
3
91.45
-2.56% include/sw/universal/verification/correctly_rounded_reference.hpp
Jobs
ID Job ID Ran Files Coverage
1 43767352162.1 15 Nov 2026 10:25PM UTC 894
89.15
GitHub Action Run
Source Files on build 43767352162
  • Tree
  • List 894
  • Changed 2
  • Source Changed 1
  • Coverage Changed 2
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #43767352162
  • 41e76fb4 on github
  • Prev Build on main (#43749674577)
  • Next Build on main (#43775712941)
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