pjfanning opened a new pull request, #1331: URL: https://github.com/apache/poi/pull/1331
Item 5 of the evaluator performance list. **Measured effect is marginal — see the numbers before deciding whether it earns its place.** ### What it does `CellCacheEntry.recurseClearCachedFormulaResults` (the walk behind `notifyUpdateCell`) called `getConsumingCells()` at every node, copying the consumer set to an array, because each consumer's `clearFormulaEntry()` removes it from the set being iterated. The walk now drains the set instead: `FormulaCellCacheEntrySet.peekAny()` returns some entry, clearing it removes it, repeat until the set is empty. A scan cursor carries on from where the last entry was found, so draining stays linear despite the holes removals leave in the open-addressing array (and survives the shrink re-hash). `containsPeeked` is an O(1) guard that turns a broken invariant (a consumer that does not list this cell as an input) into an `IllegalStateException` instead of an endless loop. The listener variant of the walk (test/debug only) keeps the array because `sortDependentCachedValues` needs one. A consumer already cleared by a deeper recursion has left the set by the time the loop would reach it, so it is skipped rather than cleared a second time — same end state as before. ### Numbers `HSSFFormulaReevaluationBenchmark` (`-Pjmh.profilers=gc`), clean iterations (both runs have GC-boundary outliers): | rows | alloc/op trunk | alloc/op this | time | |---|---|---|---| | 100 | 308.5–308.8 KB | 307.1–307.3 KB | ~1.04 ms, unchanged | | 200 | 613–615 KB | 611–613 KB | ~1.8 ms, unchanged | About 0.5% less allocation, no measurable time difference. The copy is one small array per invalidated node and the `clearFormulaEntry` work at that node (unregistering from every input) dominates it. It would matter more for a cell with thousands of consumers, but even there it is one array per notification. ### Tests - `TestFormulaCellCacheEntrySet` (new): add/remove/duplicate, draining 1000 entries through the grow and shrink re-hashes with `peekAny`/`containsPeeked`, `peekAny` not removing. - `TestCellCacheEntry`: invalidation walk over a diamond with a self-consuming (circular) formula clears every dependent and every registration; 500 consumers of one cell. - Formula suites green. 🤖 Generated with [Claude Code](https://claude.com/claude-code) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
