This benchmark appends a function result to a known array, 1,000 times per round, and accumulates a checksum. It isolates the append fallback that must materialize its RHS before writing. It does not represent every PHP workload.

With PHP 8.5 and a matching PHPX/embed installation, from this directory:

php run.php 100000
php ../../bin/tpc.php project.yml --no-progress -o array_append_benchmark
./array_append_benchmark 100000

The checksum must be 200000000. Build baseline and candidate revisions into separate build directories and binary paths. Keep PHPX, PHP, compiler flags, and machine fixed. Alternate binary execution order, discard a warm-up pair, and compare medians across multiple runs; exclude compilation time.

Linux ARM64 in Docker, PHP 8.5.10 ZTS, PHPX 6a68f38, GCC -O2, no LTO; baseline TypePHP 692841a6 versus direct known-array append lowering. Nine measured runs per binary after one discarded pair, 100,000 rounds:

Elapsed time decreased by 42.54% for this append-heavy microbenchmark; all nine paired runs were faster. This is not an application-wide speedup or a comparison against native PHP. Measurements include process startup and were collected on a shared development machine. Raw times in seconds are in results-arm64.json; its count field is the number of rounds.