[timacdonald]

tl;dr; Improve caching performance by up to 55x

Problem

FileCache::has and FileCache::get both call FileCache::all, which reads the entire cache file under a lock.

A run checks every file with has AND get, so a fully cached run includes the cache file twice per analysed file.

On a project with ~4,200 files the cache file is ~800KB and each include takes ~6ms, so a fully cached run spends most of its time re-reading the cache.

This keeps the cache contents in memory after the first read. persist and flush clear them, so the next read loads the file again.

Writes still go to disk under the lock as before, so forked workers still merge their results.

Benchmarks

Laravel application with 4,179 analysed files, 11 cores, two rounds each:

Run	Before	After
Nothing cached	33.0s / 31.3s	33.6s / 35.6s
Fully cached	29.5s / 30.3s	0.55s / 0.54s

The JSON report was identical across every run. A file with missing types reported the same uncovered lines from a cold run, from the cache, and after being changed while every other file was cached.