reduxjs/reselect v5.3.0

on GitHub

This feature release adds a maxSize option to weakMapMemoize to bound cache growth, adds a new development-mode check that warns when a selector's cache grows without bound, improves memoization performance, removes the experimental unstable_autotrackMemoize, and updates our TypeScript support matrix and documentation.

Changelog

maxSize Option for weakMapMemoize

weakMapMemoize has been the default memoizer since v5.0, and its main benefit is the infinite cache size. Prior to v5, selector instances had a default cache size of 1, which meant having to create unique selector instances per component when sharing selectors that took varying arguments like IDs. weakMapMemoize memoizes based on all arguments, so it eliminated the extra setup work and just stores all cached values.

However, that behavior can also effectively turn into a memory leak depending on what state and arguments are passed in and how they're used in the UI.

weakMapMemoize now accepts a maxSize option that bounds this growth:

const getVisibleItems = weakMapMemoize(
  (items, from, to) => items.slice(from, to),
  { maxSize: 100 }
)

This shares the same maxSize option name as the earlier lruMemoize, but has different behavior. Rather than an LRU eviction, this is a "generational" swap (kind of like a double buffer). When the cache hits its max size, it's swapped out with an empty version and starts over at 0 entries. So, there's effectively at most 2 * maxSize items in memory at any time. If maxSize is not enabled, there's no additional logic or performance overhead. If you do need LRU-style behavior, use lruMemoize instead.

Note that bounding a createSelector selector requires passing maxSize in both memoizeOptions and argsMemoizeOptions, since the two memoization levels have separate caches. See the maxSize docs for details.

Thanks to @veksa for proposing this in PR #761 and providing memory test infrastructure that helped verify this behavior.

New cacheSizeCheck Dev-Mode Check

We've also added an additional dev-mode check alongside the existing inputStabilityCheck and identityFunctionCheck. If a memoized function has accumulated over 1000 values for the same args, it logs a warning with the function name and stack trace. By default this runs once per function. Unlike the other two checks, it can only be configured globally, via setGlobalDevModeChecks({ cacheSizeCheck: 'always' | 'once' | 'never' })

Performance Improvements

We've revamped our own performance benchmark suite to give better results with more precision and less noise. That's helped verify some additional performance improvements.

• weakMapMemoize now returns early on a cache hit instead of continuing through bookkeeping
• lruMemoize's cache lookup was simplified to eliminate unnecessary allocations

These are small wins on already-fast paths, but did show modest improvements.

Removal of unstable_autotrackMemoize

We've removed the experimental unstable_autotrackMemoize export. It was added in v5.0 as an experiment in Glimmer-style dependency tracking, never left unstable_, and as far as we can tell never saw real adoption. If you were using it, switch to weakMapMemoize (the default) or lruMemoize.

TypeScript Support

Our TypeScript support matrix is now 5.6 and up, matching DefinitelyTyped, and CI now tests against TS 6.0.

We've documented on the selector fields that a full cache reset requires clearing both memoization levels: selector.clearCache() plus selector.memoizedResultFunc.clearCache().

We improved types handling in cases where TS strict mode is off (but please migrate to strict behavior as soon as possible!)

Docs Improvements

The API docs got a structural overhaul: every API page now follows a consistent "API Reference / Usage Guide" layout, the dev-checks page was rewritten, and the docs reflect that weakMapMemoize is the default memoizer since v5. We've also added some additional usage guidance as well.

What's Changed

• Additional publishing hardening by @markerikson in #755
• Bump arethetypeswrong CLI to 0.18.3 to fix fflate tarball crash by @veksa in #764
• Don't widen an all-optional selector result to any by @veksa in #776
• Fix assorted memoization correctness issues by @markerikson in #777
• test: add benchmarks for cache fill cost and retained cache size by @veksa in #769
• test: paired hot-path benchmark for createSelector by @veksa in #770
• perf(createSelector): gather input selector results without extra allocations by @veksa in #772
• perf(createSelector): stop allocating dev-check bookkeeping on every call by @veksa in #773
• perf(weakMapMemoize): return on a cache hit instead of rewriting the node by @veksa in #771
• Improve perf harness by @markerikson in #778
• Optimize lruMemoize cache lookups by @markerikson in #779
• Add regression test for resultEqualityCheck receiving a cleared WeakRef by @veksa in #763
• Update test infra and add additional memory tests by @markerikson in #780
• Add weakMap cache size dev check and revamp docs by @markerikson in #782
• Add a maxSize option for weakMapMemoize by @markerikson in #783
• docs: clarify how to fully clear a selector's cache (#569) by @veksa in #760
• Remove experimental autoTrackMemoize by @markerikson in #784
• Add docs note on complex selectors and state by @markerikson in #785

Full Changelog: v5.2.0...v5.3.0