reflex-dev/reflex

reflex

Performance History

Latest Results

Document that 0.9 and 0.10 instances cannot share a state store Add a section to the 0.10 upgrade guide saying that a 0.10 instance loads state saved by 0.9 but a 0.9 instance discards state saved by 0.10, so a rolling deploy that runs both versions, or a rollback against the same Redis or disk store, resets the sessions that reach the older instance. The breaking-change fragment and the Self Hosting note are in #7494. Refs #7470 #7474 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SjCrbdguMkfe4bhbaREiwo

claude/docs-0.10-migration

14 minutes ago

Declare state stores shared by 0.9 and 0.10 instances unsupported Drop the change that wrote the 0.9 schema hash next to each pickled state so 0.9 and 0.10.0a1 workers could load state saved by this release, and restore reflex/state.py and its tests to main. Instead, a `breaking` news fragment and the Self Hosting docs state that instances of 0.9 and 0.10 cannot share a Redis or disk state store: 0.10 still loads state saved by 0.9, but a 0.9 instance discards state saved by 0.10, so a rolling deploy that runs both versions, or a rollback against the same store, resets the sessions that reach the older instance. The field-descriptors fragment no longer claims that states saved by this release load in workers of the previous one, which has not held since the names-and-types schema hash. Refs #7470 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SjCrbdguMkfe4bhbaREiwo

claude/n004-pickle-compat

16 minutes ago

Bound replaced-default history and cache the mangled-name prefixes Address the review round on the class-assignment fixes. Field._assign_default kept every replaced default for patch/restore, so a state that reassigns a class default at runtime grew the list without bound. The history is now a deque of at most _MAX_REPLACED_DEFAULTS (16) entries, dropping the oldest: undoing still unwinds nested patches, and repeated plain assignments stay bounded. The attribute name is one constant, _REPLACED_DEFAULTS_ATTR, used by _RESERVED_FIELD_ATTRS, _assign_default and _restore_default. ComponentState.create already copies fields per generated class without their history; a unit test now pins that. Fix a dev-mode performance regression from the mangled-name guard: _is_plain_private_name rebuilt the prefix of every class in the MRO on each call, so the first write of each name on a fresh state class cost about 2.5 us instead of about 60 ns, and test_state_tree_delta ran 24-29% slower on CodSpeed. Names without a double underscore now return at once, and the prefixes are cached per class with _cache_per_class (class names are fixed once __init_subclass__ renames local states). The base vars docs no longer claim that a declared default_factory producing browser storage keeps its storage options on a plain string assignment; it does not. The browser test now also covers SessionStorage: the value is stored under its configured name, survives a same-tab reload, and a new tab starts from the assigned default. Refs #7471 #7472 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SjCrbdguMkfe4bhbaREiwo

claude/class-assignment

16 minutes ago

Delete tests/units/test_optional_dependencies.py

claude/db-extra-greenlet

41 minutes ago

Sort the sidebar imports Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SjCrbdguMkfe4bhbaREiwo

claude/docs-0.10-migration

42 minutes ago

Use an absolute sidebar import and split the background-task example The docs app requires absolute imports from the project root, and the upgrade guide's example raised at its first line on 0.10 before reaching the working call, so the old and new versions are now separate snippets. Refs #7474 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SjCrbdguMkfe4bhbaREiwo

claude/docs-0.10-migration

43 minutes ago

Point reflex component at the component template The custom components CLI was removed in 0.10, so every `reflex component` subcommand failed with "No such command 'component'. Did you mean 'compile'?" and no hint about where the workflow moved. Register a hidden `component` shim that accepts any subcommand and flags, logs that the command was removed in Reflex 0.10, points at the wrapping React docs and https://github.com/reflex-dev/component-template, and exits 1. `reflex component --help` prints the same pointer. The command stays out of `reflex --help`, and the plural typo `reflex components` now suggests it. Fixes #7490 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CEngePWmkuAQ31xWPb1Q9L

claude/beautiful-edison-jtk7eu

47 minutes ago

Serialize non-class field types in the previous-release schema helper The helper that recomputes the hash 0.9 workers check mirrored only the class branch of the old _serialize_type, so a state with a generic or union annotation would raise instead of being checked. Mirror the other branch too, as the 0.9.12 implementation did. Refs #7470 Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SjCrbdguMkfe4bhbaREiwo

claude/n004-pickle-compat

48 minutes ago

Latest Branches

+4%

Add the 0.10 upgrade guide, the missing sqlmodel release note, and a BackendVarFormatError hint#7496

25 minutes ago

a0b6aef

claude/docs-0.10-migration

+10%

Declare state stores shared by 0.9 and 0.10 instances unsupported#7494

25 minutes ago

c6e7cd9

claude/n004-pickle-compat

+16%

Fix class-level State attribute regressions from #7461 / #7465 (patch restore, storage defaults, mangled-name guard)#7495

34 minutes ago

c7e6243

claude/class-assignment

Home Terms Privacy Docs