Split str.format below the cognitive-complexity threshold (0061)

[rmorozov]

Requested by Ruslan · project thread

Before: StrOps<CharT>::Format in src/value_methods.cpp implements "...".format(...). One function scanned the format string, parsed each replacement field, looked up the argument, applied the .attr and [key] accessors and the !r/!s conversion. Its cognitive complexity was 88 (threshold 25), and it was kept under a NOLINT marker.

After: Format only scans for braces and hands each {...} to ReplaceField. ReplaceField uses two helpers: LookupArg, which resolves a position or keyword and enforces Python's no-mixing rule for {} and {0} through a FieldNumbering, and ApplyAccessors. The brace scan now uses a single doubled flag for both {{ and }}. Output, error messages and which error is reported first are all unchanged. 28 complexity markers remain. Part of docs/tasks/0061.

Roles and rounds: main session plus one verifier round, which found nothing blocking.

• The verifier rendered about 560k format strings through the parent and through this commit, narrow and wide, with six argument sets each. The strings were 200k structured or random, 250k random, and an exhaustive set of every string up to length 5 over {}!:.[]0rx. The output was byte-identical, every error path included.
• An ASan+UBSan build was clean, and so were clang-tidy and clang-format.

Red pushes: 0 so far.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LJv5o6nbcd16qaXtwhoNVG

---

Generated by Claude Code