sequenceDiagram
  participant Caller as Python Caller
  participant PyFloat as PyFloat.fromhex
  participant FloatOps as float_ops::from_hex
  participant Finish as finish_hex

  Caller->>PyFloat: fromhex(string)
  PyFloat->>FloatOps: from_hex(string.as_str())
  FloatOps->>FloatOps: parse whitespace, sign, inf/nan, 0x prefix, digits, exponent
  FloatOps->>Finish: finish_hex(remaining, sign, value)
  Finish-->>FloatOps: Ok(f64) or HexFloatError
  FloatOps-->>PyFloat: Result<f64, HexFloatError>
  alt Overflow
    PyFloat-->>Caller: raise OverflowError
  else TooLong / Invalid
    PyFloat-->>Caller: raise ValueError
  else Ok
    PyFloat-->>Caller: return float
  end

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands.}

📦 Library Dependencies

The following Lib/ modules were modified. Here are their dependencies:

[x] test: cpython/Lib/test/test_float.py (TODO: 3)
[x] test: cpython/Lib/test/test_strtod.py (TODO: 2)

dependencies:

dependent tests: (no tests depend on float)

Legend:

• [+] path exists in CPython
• [x] up-to-date, [ ] outdated

Thanks for the review — addressed in 51b3c1a:

• byte_at (was hex_byte_at) now returns Option<u8> (bytes.get(i).copied()); call sites use == Some(..) / matches!(.., Some(b'p' | b'P')), which also folds in the matches! suggestions.
• hex_from_char now returns Option<u8> and is const fn; the scans go through a small hex_digit_at peek-and-decode helper.
• parse_inf_or_nan now returns Option<(f64, usize)> (no more -1.0 sentinel).
• whitespace helper — you're right that it already existed: the copy is is_py_ascii_whitespace in sre_engine. I hoisted a shared pub const fn is_py_ascii_whitespace into common::str and use it here. I deliberately left sre_engine's private copy in place: sre_engine is a standalone low-level crate (it only depends on wtf8), and making it depend on common would pull the whole runtime (malachite-bigint, icu, …) into a published regex crate. Happy to have sre_engine delegate to the shared one instead if you'd prefer that tradeoff.
• ldexp is now const fn; the case_insensitive_match doc wording is fixed.
• Dropped the redundant test_ prefix on the unit tests (this was the clippy::redundant_test_prefix Lint failure).

— commented by Claude

Update on the whitespace helper (after some more thought on where it belongs): instead of common::str, the shared is_py_ascii_whitespace now lives in rustpython-wtf8 and is used by both common::float_ops and sre_engine. wtf8 is the string crate that both already depend on, so this adds zero new dependency edges, and sre_engine's previous private copy is removed in favor of it. (rustpython-literal was considered but reverted, since it would have added a new dependency to sre_engine.) See bd187bf91.

For the record, on two related questions that came up: the hex-float parser can't be borrowed from ruff (Python source has no hex-float literals, so ruff's number lexer only handles hex/oct/bin integers + decimal floats — the only fromhex in ruff is vendored typeshed stubs), and ldexp isn't reused from pymath because pymath only exposes ldexp_bigint(f64, BigInt), which would force a BigInt allocation per call for what is an i32 exponent here — hence the small self-contained scalbn.

— commented by Claude