Slate v2 History Perf Batch

Goal

Close the missing withHistory(createEditor()) perf lane for strict RC acceptance and stop hand-waving about history cost.

Kept Work

• added the compare lane: history.mjs
• wired the command: bun run bench:history:compare:local
• restored legacy-style typing merge heuristics in with-history.ts

Result

Latest read at 5000 blocks, 20 typed characters, 200 inserted fragment blocks:

• current typing undo 29.71ms
• current typing redo 20.53ms
• current fragment undo 27.21ms
• current fragment redo 42.95ms

Legacy:

• typing undo 0.36ms
• typing redo 0.49ms
• fragment undo 1.92ms
• fragment redo 11.18ms

So yes, current slate-history is still slower than legacy.

But the real blocker-level disaster was still fixed first:

• before restoring merge heuristics, typing-burst undo/redo was roughly ~490ms
• after the fix, the same lane is back in the low 20-40ms band

That is the difference between “RC blocker” and “strict-acceptance proof row that stays slower than legacy but still comfortably fast.”

Verdict

Keep the lane.

Do not reopen RC on this alone.

Reason:

• this is a headless compare lane, not a user-facing browser regression
• the remaining cost is real, but still lands in low tens of ms on a 5000 block document
• the deep-interview blocker policy already said slower-than-legacy can stay non-blocking when it is still comfortably fast in practice