jp-string-plus-in-loop — removed 2026-10-23 (Asia/Shanghai)
Status: removed. The matcher saw += of a string literal or String.fromCharCode and told you to switch to push then join. That rewrite does not have about a 2× gain on those shapes. The matcher is deleted. The slug is never reused.
How measured
examples/bench/evidence_jp005_jp006.js. Nine rounds after three warm-up runs; median. Both sides build the same string and read its length. Node v24.18.0, V8 13.6.233.17-node.50. JSONL: jp-string-plus-in-loop.jsonl.
ratio is += time / join time. About 2 would mean join is twice as fast. Under 1 means join is slower.
| role | case | median_ms | ops | ratio |
|---|---|---|---|---|
| before | s += "\n", 100 chars × 200 | 0.194 | 20000 | 1.534 |
| after | push + join | 0.126 | 20000 | 1 |
| before | s += "\n", 22000 chars × 20 | 1.648 | 440000 | 0.598 |
| after | push + join | 2.756 | 440000 | 1 |
| before | s += "\n", 44000 chars × 10 | 1.459 | 440000 | 0.459 |
| after | push + join | 3.176 | 440000 | 1 |
| before | String.fromCharCode, 22000 × 10 | 0.893 | 220000 | 0.337 |
| after | push + join | 2.647 | 220000 | 1 |
| before | String.fromCharCode, 44000 × 5 | 0.688 | 220000 | 0.197 |
| after | push + join | 3.494 | 220000 | 1 |
| before | 64-char literal × 700 × 40 | 0.070 | 28000 | 0.310 |
| after | push + join | 0.226 | 28000 | 1 |
| before | obj.s += "\n", 22000 × 10 | 0.759 | 220000 | 0.418 |
| after | push + join | 1.816 | 220000 | 1 |
| before | s += ch (the matcher did not see this) | 1.178 | 440000 | 0.417 |
| after | push + join | 2.823 | 440000 | 1 |
The only case where join was faster is 100 newlines, at 1.53×, under the gate. At 22000 and 44000 characters, join is slower. V8 already builds the string for += in a loop. push then join also allocates the pointer array.
Fixtures
Historical bad/good fixtures live under _removed/fixtures/. They are not part of the smoke run.