cx-string-append-unreserved — string += grow vs reserve (-O2)
这篇还没有中文，下面是英文原文。
Claim
Measured on this machine (2026-10-22 re-run): primary hot case string_plus_grow (char/"x" append) ratio before/after ~1.09× (~1×). Earlier fill ~0.97× reverse. Deep-dive: single-char append often ties or loses to reserve at -O2; 64-byte chunk append sees reserve win ~1.8–1.9×. Default stays hint. Do not escalate to error.
How measured
- Harness:
packages/cppperf/examples/bench/evidence_cx.cpp - Opt:
O2 - Deep-dive chunk arm: separate script (case
CX002_chunk64), not shipped in this repo - Raw lines:
cx-string-append-unreserved.jsonl
Runtime (this evidence file)
| Field | Value |
|---|---|
| When | 2026-10-22 Asia/Shanghai (machine clock; JSONL ts is UTC) |
| Toolchain | g++ (Debian 14.2.0-19) 14.2.0 |
| Opt | O2 |
Median after (string_reserve_plus) | 1.162903 ms (ops=200000, ratio=1.000000) |
Median before (string_plus_grow) | 1.263970 ms (ops=200000, ratio=1.086909) |
| Primary ratio before/after | ~1.09× (string_plus_grow) |
Lint mapping
cppperf-lint cx-string-append-unreserved default stays hint. Escalate to error recommended: no. Lint message / docs narrowed to known final size or multi-byte chunk append (2026-10-22 deep-dive land).
Limits
- Fixture uses 1-char append — weak for the smell. Prefer documenting chunk/known-final-size cases.
- The null/reverse deep-dive judged char
+=scope-narrow / smell-false; only the chunk arm is kept. - Chunk arm is the intended win; char += is not a speed gate.
Memory (2026-10-22)
Retained capacity_bytes: char grow 245760 vs reserve 200000 (1.23×); chunk64 grow 4194304 vs reserve 3200000 (1.31×). Chunk arm is the clearer footprint story. cx-string-append-unreserved.mem.jsonl.