stillwater-sc / universal / 25980502833

84%

master: 84%

push

github

feat: API parity for decimal string parse in extended-precision FP family (Phase C of #835) (#847)

* feat(dd, qd, dd_cascade, td_cascade, qd_cascade): API parity for decimal string parse (Phase C of #835)

Phase C scope for the extended-precision FP family is API parity, not
a precision rewrite. Unlike posit/cfloat (where stod truncated to 53
bits up front), the dd/qd/cascade parse functions already accumulate
digits in their own high-precision arithmetic, preserving their full
mantissa width all the way through.

What was missing was the API surface:

1. NaN / Inf / Infinity token routing
   The digit-accumulation loop in each type rejected any alphabetic
   character outright, so parse("nan") / parse("inf") returned false.
   Each parse() now detects nan / inf / infinity (any case, optional
   sign, leading whitespace tolerated) before the digit path and
   routes to setnan(NAN_TYPE_QUIET) / setinf(negative).

2. operator>> failbit on parse failure
   The dd / qd extractors emitted std::cerr but did not setstate(failbit),
   so callers had no programmatic signal. The cascade types had no
   operator>> at all; added one with the standard cerr-plus-failbit
   pattern, matching the symmetry of posit / cfloat / integer / fixpnt.

The digit-accumulation core, the pown(10, e) exponent step, and all
existing semantics are unchanged. The residual ULP-level error in
pown(10, e) for large |e| remains a known limitation, deferred to a
future phase if a d2b-based replacement is warranted.

Tests:
- static/highprecision/dd/conversion/string_parse.cpp
- static/highprecision/qd/conversion/string_parse.cpp
- static/highprecision/dd_cascade/conversion/string_parse.cpp
- static/highprecision/td_cascade/conversion/string_parse.cpp
- static/highprecision/qd_cascade/conversion/string_parse.cpp

Each covers canonical decimals, nan/inf token routing (all four
case-and-sign spellings), and operator>> failbit on a bad token,
using the ReportTestSuite pattern.

Verified on ... (continued)

104 of 109 new or added lines in 6 files covered. (95.41%)

46383 of 55166 relevant lines covered (84.08%)

6424114.9 hits per line