rs-string-plus-in-loop — String += / push_str in loop
Claim
Plain s += "…" / push_str on a single growing String is amortized ~1× vs String::with_capacity under rustc -C opt-level=3 (geometric growth already avoids Lua/..-style copies). Tiny builders can show a small capacity win when many fresh String::new() starts are repeated. This rule does not cover format! rebind accumulation — that is rs-format-rebind-hot.
How measured
- Harness:
packages/rustperf/examples/bench/evidence_rs002.rs(pairs A/B/D) - Build:
rustc -C opt-level=3 evidence_rs002.rs -o out/evidence_rs002 - Timer:
std::time::Instantwall clock; warm=3; median of 9;std::hint::black_box - Pair A (matches fixture
rustperf-lint/fixtures/bad/rs-string-plus-in-loop_string_plus.rs/ capacity intent): N=500000- before:
s += "x"fromString::new() - after:
String::with_capacity(N)+push_str("x")
- before:
- Pair B (fixed 32-byte chunks): N=100000
- before:
push_str(chunk)without capacity - after:
with_capacity(N*32)+push_str
- before:
- Pair D (small N=100, ×20000 outer reps for timer floor): expect modest / ~1–2×
- Raw lines:
rs-string-plus-in-loop.jsonl(format! pair moved tors-format-rebind-hot.jsonl)
Runtime (this evidence file)
| Field | Value |
|---|---|
| When | 2026-10-22 Asia/Shanghai (machine clock; JSONL ts is UTC) |
| rustc | 1.85.1 |
| Opt | opt-level=3 (-C opt-level=3) |
Median before (plus_eq_loop) | 0.345 ms |
Median after (plus_with_capacity) | 0.344 ms |
| ratio plus before/after | ~1.00× |
Median before (chunk32_push_str_loop) | 0.283 ms |
Median after (chunk32_with_capacity) | 0.270 ms |
| ratio chunk32 before/after | ~1.05× |
Median before (small_n_plus_eq) | 3.238 ms |
Median after (small_n_with_capacity) | 1.588 ms |
| ratio small N | ~2.04× |
Direction: fixture-shaped += / push_str stays ~1×. Absolute ms and ratios are N/machine/opt-sensitive. Re-run the harness after rustc or machine changes.
Lint mapping
rustperf-lint rs-string-plus-in-loop flags += / .push_str( inside a real for/while/loop window (reverse-opt skips tiny seps, help/error formatters, clap_complete shells — see G3 triage). Default severity hint: JSONL exists but the flagged smell is amortized ~1×. Do not CI-error clap-like builders. Hot format!("{s}…") / s = format!(…) rebind is rs-format-rebind-hot (separate detector + evidence).
Limits
- Microbench; not an app profile.
- Rust
Stringgrowth is geometric — do not equate rs-string-plus-in-loop with Lua../ lu-string-concat-hot quadratic copies for plainpush_str. - Pair D multiplies small builders for timer resolution;
opsis chars per builder (100), not total appends. - Does not measure
Writetrait objects,itoa, or SSO. - Historical note: early G3 was OK-idiom-heavy; kit later flipped corpus-ready (see docs/corpus.md). rs-println-hot has since been measured (hint).