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

stillwater-sc / universal / 39158719339
89%
master: 84%

Build:
Build:
LAST BUILD BRANCH: main
DEFAULT BRANCH: master
Ran 18 Oct 2026 07:38PM UTC
Jobs 1
Files 752
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

18 Oct 2026 07:25PM UTC coverage: 87.447%. Remained the same
39158719339

push

github

web-flow
fix(efloat): parse rounds a literal of any length to the full precision (#1740)

parse() capped its target below BigBits - (3*|E| + digits), a ceiling that
fell as the literal got longer, and refused a literal whose digits and
exponent did not fit BigBits. assign() discards parse()'s result, so a
300-digit constant became 64 bits and a 310-digit one became 0, silently.

decimal_to_binary::convert already widens its intermediates as a literal
needs (#1504), so the pre-checks are gone and the target is the value's
precision. The conversion runs at target + 64 guard bits and rounds once to
nearest-even. Digits past the first 1 + (target + 64) log10(2) significant
ones are dropped and kept as a sticky bit, so the cost no longer grows with
the literal; the one guard pattern where they can decide the rounding,
0111...1, retries with twice the digits.

Past the converter's widest working width (efloat<>'s default 32768 bits is
beyond it) the target is lowered per literal to the most it can reach: about
32700 bits for a short literal, about 18500 for a 2500-digit one. A
well-formed literal not even one bit of which fits (an exponent past about
+/-13000) returns false and leaves a NaN; malformed text still leaves 0.

Tests: long_literals.cpp checks parse() and assign() bit for bit against
exact rational references (tools/generators/efloat_parse_gen.py) at 608 and
2048 bits, including exact midpoints and their neighbours, and efloat<> at
its default precision. 86 of the 88 vector cases fail on main.

Resolves #1713

BEHAVIOUR-CHANGE: efloat parse()/assign() of long literals, or at precisions above ~1900 bits (including efloat<>'s default), now return the correctly rounded value instead of a truncated one or 0; parse("1e1000") with the default budget succeeds; an unreachable literal such as 1e100000 leaves NaN instead of 0

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

92 of 98 new or added lines in 1 file covered. (93.88%)

4 existing lines in 2 files now uncovered.

51279 of 58640 relevant lines covered (87.45%)

11026091.92 hits per line

Uncovered Changes

Lines Coverage ∆ File
6
89.66
0.29% include/sw/universal/number/efloat/efloat_impl.hpp

Coverage Regressions

Lines Coverage ∆ File
2
96.91
-0.13% include/sw/universal/number/cfloat/cfloat_impl.hpp
2
82.83
-0.3% include/sw/universal/number/qd/qd_impl.hpp
Jobs
ID Job ID Ran Files Coverage
1 39158719339.1 18 Oct 2026 07:38PM UTC 752
87.45
GitHub Action Run
Source Files on build 39158719339
  • Tree
  • List 752
  • Changed 3
  • Source Changed 1
  • Coverage Changed 3
Coverage ∆ File Lines Relevant Covered Missed Hits/Line
  • Back to Repo
  • Github Actions Build #39158719339
  • 5b7cc5d0 on github
  • Prev Build on main (#39153583842)
  • Next Build on main (#39162537678)
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