Storybook 10.5 module-graph perf reproduction

A reproduction for a performance regression introduced in storybook 10.5. The ModuleGraphEngine rebuilds and broadcasts the entire module-graph reverse index on every file-watcher event.

Problems demonstrated

1. Unconditional rebuilds: ModuleGraphEngine.mirrorUpdate() computes the delta of changed files (bumpedStoryFiles). But it then re-serializes the whole index anyway. handleFileChange() also discards IncrementalPatcher.patch()'s return, which already knows the dependency set was unchanged.
2. Unrelated files trigger the rebuild: A file absent from the graph skips the change check entirely, so writing a log file costs as much as editing a story.

Setup

npm install

Optionally, instrument the cause:

node instrument.mjs      # logs each index rebuild and its size
node instrument.mjs --revert

Reproduce

npm run storybook        # port 6008, no browser needed

Wait for startup to finish. Then note the dev-server process's cumulative CPU (ps -o time= -p <pid>), and:

echo "" >> src/stories/Comp0.stories.jsx    # a real source edit
echo "noise" >> unrelated.log               # a file the graph has never seen

Check the CPU delta after each. Both cost the same.

Measured here (400 stories x 1200 modules => 480,800 index pairs, node 22, macOS arm64, 16 cores):

story save:           ~2.66s CPU, 1 rebuild
unrelated .log write: ~2.53s CPU, 1 rebuild

Confirming it's a 10.5 regression

Downgrade Storybook, leaving everything else alone:

npm install --save-dev storybook@10.4.6 @storybook/react-vite@10.4.6

Re-run the same two tests. The cost disappears:

                      10.5.5    10.4.6
story save:           ~2.74s     ~0.18s
unrelated .log write: ~2.71s     ~0.01s

Scaling

The performance gets worse with a larger module graph. If it's useful to see how performance degrades with module graph size, one can generate a larger repro with the generate script:

STORIES=2200 MODULES=1500 node generate.mjs

(Performance cost also seems to depend on module graph topology, but I left that out of this repro for simplicity. I can expand more for ya if it's actually useful data.)