Module stdlib 

The numeric core of C’s strtod for C99 hex floats, shared by every strtod port in the workspace (calc::engine::strtod, the CALC literal scanner, and epics-ca-rs’s estdlib, the env-knob parser).

Both ports used to build the significand in an f64 (mant = mant * 16.0 + digit) and then scale it with mant * 2.0f64.powi(exp). That composition is wrong twice over, and the whole subnormal range paid for it:

• powi with a negative exponent evaluates 1.0 / 2^-exp, and 2^1074 is already infinite — so every exponent below about -1023 scaled the significand by 1/inf == 0. 0x1p-1074 came back as an underflow to zero where glibc returns the exact smallest subnormal.
• ERANGE was then guessed back from the result: any subnormal was reported as an ERANGE overflow. glibc raises ERANGE only when the exact value is tiny AND inexact, so 0x1p-1023 — an exactly representable subnormal — leaves errno clear.

HexSignificand keeps the digits as an exact integer m * 2^e2 with a sticky bit for anything that falls off the bottom, so the conversion to f64 is a SINGLE correctly-rounded step (ties to even) and knows precisely whether it was inexact. Every row of the boundary table in this module’s tests was probed against the compiled glibc strtod on this platform.

Structs

HexSignificand

A C99 hex float’s significand, accumulated exactly.

Enums

ParseDoubleError

The epicsParseDouble failure codes (epicsStdlib.h).

Functions

epics_parse_double

C epicsParseDouble(str, to, NULL) (epicsStdlib.c:149-176): skip leading whitespace, run strtod, reject ERANGE, skip trailing whitespace, reject anything left over.

epics_scan_double

C epicsScanDouble (epicsStdlib.h:203) — epicsParseDouble with the status collapsed to a boolean.