Benchmarks¶

PHPantom is benchmarked on every commit to track performance regressions. All numbers below were measured on a production Laravel codebase: 5.1K PHP files (389k lines) and 1.3K Blade templates (88k lines), with 27k vendor files (1.7M lines).

Headline Numbers¶

Metric	PHPantom	Intelephense	PHP Tools	Phpactor	PHPStorm
Cold start, CPU time	17 s	16 s	36 s	3 min 18 s	17 min 55 s
Cold start, wall clock	2 s	11 s	10 s	3 min 17 s	1 m 7 s
Warm start, CPU time	17 s	4 s	36 s	3 s	3 min 23 s
Warm start, wall clock	2 s	1 s	10 s	4 s	5 s
RAM, cold start	578 MB	766 MB	594 MB	467 MB	1 GB
RAM, warm start	578 MB	759 MB	594 MB	60 MB	2 GB
Disk cache	0	51 MB	0	2.3 GB	379 MB

A cold start is the first index with no disk cache; a warm start reuses the cache written by a previous session. CPU time is the total consumed until full type intelligence is available; wall-clock times are approximate. RAM is the steady resident memory once indexing has finished, not the peak reached during the index itself. Intelephense and Phpactor do not analyse Blade templates or Laravel's runtime magic, which makes this codebase cheaper for them than the numbers alone suggest.

The codebase was stripped to its PHP and Blade files before each run, with the .git directory and all JS/CSS assets removed, so every tool indexes the same PHP-only surface. PHPStorm's baseline IDE startup time was measured separately and subtracted from its cold and warm start figures above, to keep the comparison focused on PHP indexing rather than general editor overhead.

Live Charts¶

Latency and memory usage are tracked on every commit and plotted over time. These charts are useful for catching regressions and observing trends across releases.

Latency Benchmarks -- completion response time per commit
Memory Benchmarks -- resident memory after full indexing

What We Measure¶

Latency benchmarks run cargo bench on the completion engine, measuring end-to-end response time for completion requests against real-world fixtures. Results are tracked with github-action-benchmark.

Memory benchmarks measure peak resident memory (RSS) of phpantom_lsp after fully indexing the benchmark project. THP (Transparent Huge Pages) is disabled during measurement for consistent results.