reflex-dev/reflex - CodSpeed

reflex

Blog Docs Changelog

Performance History

Latest Results

GitHub Actions run

1 second ago

fix(state): avoid leaking filtered async vars (#7410) * fix(state): avoid leaking filtered async vars * refactor(state): resolve async vars via descriptors

main

14 minutes ago

Merge branch 'main' into codex/fix-initial-token-guidance

codex/fix-initial-token-guidance

28 minutes ago

test: skip unconfigured Postgres tests before importing psycopg (#7502)

main

30 minutes ago

bench: drop the warm metadata walk from the hydration benchmarks (#7503) test_hydration_metadata re-read the cached name, full name, parent and root of every class in a 20- and a 200-substate tree. It came in with #7064 to show the per-class metadata cache replacing the lru_cache whose 128-entry capacity the 200-class tree thrashed. With that cache in place both sizes measure the same cache-hit path, and that path is almost entirely CPython's type attribute cache and call specialization rather than reflex code. That makes it flip on CodSpeed with no related change: [200] alternates between 1.7 and 1.8 ms (-6.9% on #7502, -7.7% on #7501), [20] between 207.6 and 216.7 us (-4.2% on #7410 and #7434, +4.3% on the changelog-only #7445), while every other benchmark reports untouched. Reproduced under callgrind with the same pytest-codspeed instrumentation CI uses. Two runs of the module alone are byte-identical: 1,420,889 Ir for [200]. Running test_state_access.py or test_state_delta.py first in the same process gives 1,363,413 and 1,306,760 Ir (-4.0%, -8.0%) for the unchanged function, while test_hydration_snapshot[200] moves about 1% (5,335,449 to 5,392,507). The count depends on what the process did before the benchmark, not on the code it measures, so no restructuring of the body would settle it. test_hydration_snapshot stays: it walks the same metadata on its way to serializing the tree, so a regression in the metadata path still shows up there, diluted by real work enough to sit under the threshold when nothing changed. The two removed benchmarks show as skipped on CodSpeed until they are archived in the project settings. Claude-Session: https://claude.ai/code/session_011LZPUJkHZur6LUinHeFpZT Co-authored-by: Claude <noreply@anthropic.com>

main

33 minutes ago

test: accept completed fullstack launchers during shutdown (#7501)

main

39 minutes ago

test(workflow): make rate-limit burst and refill checks deterministic (#7433)

main

41 minutes ago

perf: skip the code-point copy for strings without surrogates StringVar length, indexing and slicing now call pyStrLength, pyStrAt and pyStrSlice in state.js. A string without surrogates has one code unit per code point, so the helpers use the native string methods and allocate nothing; only strings with surrogates go through Array.from. Steps other than None keep the array slice path through pyStrChars. A None string counts as empty for all four helpers.

harsh21234i:fix/7481-unicode-string-slicing

2 hours ago

Latest Branches

0%

fix(hosting): guide recovery from initial token rejection#7434

35 minutes ago

509c93e

codex/fix-initial-token-guidance

0%

Fix Unicode-aware StringVar operations#7518

2 hours ago

42e957f

harsh21234i:fix/7481-unicode-string-slicing

+11%

perf: fork the dev backend reload worker instead of using a forkserver#7113

2 hours ago

3bb4fe9

FarhanAliRaza:farhan/dev-fork-workers

© 2026 CodSpeed Technology

Home Terms Privacy Docs