stillwater-sc / universal / 32429280776

85%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

fix(elreal): bound el_api_math's constants -- UBSan timeout on main (#1365)

* fix(elreal): bound el_api_math's constants -- UBSan timeout on main

main is red. #1362's UBSan job failed, and it is NOT a sanitizer diagnostic:
889 of 890 tests passed, zero runtime errors reported, and the single failure is
el_api_math hitting ctest's 300s per-test limit.

Cause is the same accidental coupling that hit el_math_sqrt. The test checks the
named constants against DOUBLE literals -- about 17 significant digits is the
whole question -- but calls elreal_pi(), elreal_e() and the rest at each
generator's own default depth: 16, and 32 for e. That was cheap while the
refinement floors stopped the series early. With the floors gone (#1051/#1362) a
default-depth constant refines to hundreds of digits, and the test went from 34s
to 114s at -O0. UBSan instrumentation on top of that clears 300s.

Measured at -O0: 34.2s before #1362, 113.9s after, 20.8s now -- below the
original baseline, so there is margin under instrumentation rather than a
borderline pass.

Two changes:

- mathlib.hpp: the ten elreal_* constant accessors take an optional depth.
  0 means "the generator's own default", so every existing caller is unaffected;
  verified directly -- elreal_pi() and elreal_e() with no argument still agree
  with pi_zbcl<double>() and e_zbcl<double>() digit for digit (260 and 319).
  The parameter exists because those defaults are now genuinely expensive and a
  caller who needs a few digits should be able to say so.

- elastic/elreal/api/math.cpp: the constant checks pass depth 3, which covers the
  double literals they compare against with room to spare.

46/46 elreal tests pass under gcc 13.3 and clang 18.1.

Worth recording: I merged #1362 while ASan, UBSan and Coverage were still
pending, having said those were the checks that mattered for it. The Debug run I
did instead confirmed the 0-overlap invariant, which was the correctness risk I
was actually worried about -- ... (continued)

Coverage Stats

18 of 18 new or added lines in 1 file covered. (100.0%)

8 existing lines in 1 file now uncovered.

43063 of 50453 relevant lines covered (85.35%)

7146985.69 hits per line