stillwater-sc / universal / 29737938726
85%
master: 84%

LAST BUILD BRANCH: losborn_bugfix_elreal_subnormal_blocks_aug26
DEFAULT BRANCH: master
Ran
Jobs 1
Files 699
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
coverage: 84.771% (+0.03%) from 84.739%
29737938726

push

github

web-flow 
docs(ereal): mathlib.hpp no longer describes functions as Phase-0 double stubs (#1168)

* docs(ereal): mathlib.hpp no longer describes functions as Phase-0 double stubs

The header claimed "Phase 0: Stub implementations - all functions delegate to
std:: via double conversion" and listed Phases 1-4 (Taylor/Newton) as future TODO.
That has been false for some time: the ereal transcendentals are arbitrary-
precision implementations computed in ereal expansion arithmetic. Verified against
the efloat oracle -- exp, sqrt, log, sin, erf, gamma, ... agree to ~293 digits at
ereal<19> (the ~300-digit architectural ceiling of Shewchuk expansion arithmetic).
The only remaining double casts are series-term magnitude checks for loop
termination; the terms themselves are full-precision.

Comment-only change:
- rewrite the top-of-file description to state the high-precision reality and the
  architectural precision ceiling (maxlimbs<=19), pointing to efloat for unbounded
  precision
- drop the "Phase 0 stub" section labels
- replace the "Future TODO Phase 1-4" block with the actual remaining mathlib work,
  tracked under #582 (#1165 fdim/modf/rint/nearbyint, #1166 fma/scalbn/logb/ilogb,
  #1167 complex<ereal> binding)

No functional change; ereal mathlib compiles and runs unchanged on gcc and clang.

Resolves #1163
Relates to #582

* docs(ereal): qualify the precision cross-check claim (CodeRabbit #1168)

CodeRabbit rightly flagged that describing efloat as a verification "oracle"
overstates the guarantee. Reworded: exp(1) and pi (via atan) reproduce the
independently generated ~1000-digit mpmath constant literals (e, pi) to ~290
digits -- that is the ground-truth check -- and sqrt/log/sin/erf/gamma agree with
the efloat sibling implementation to the same level. Framed as corroboration
between two implementations plus known constants, not a proof against a
third-party oracle.

(Note: efloat's runtime arithmetic is fully implemented -- efloat exp(1) matches
the mpmath... (continued)

49758 of 58697 relevant lines covered (84.77%)

7034096.02 hits per line

Jobs
ID Job ID Ran Files Coverage
1 29737938726.1 699
84.77
 GitHub Action Run
Source Files on build 29737938726