* remove deprecated features and support for py3.9
* remove other deprecated stuff
* update lock file
* fix units tests
* relock poetry
* fix _replace for computed_var
* fix some merge typo
* fix typing of deploy args
* fix benchmarks.yml versions
* console.error instead of raising Exception
* fix tests
* ignore lambdas when resolving annotations
* simplify redirect logic in event.py
* more fixes
* fix unit tests again
* give back default annotations for lambdas
* fix signature check for on_submit
* remove useless stuff
* update pyi
* readd the getattr
* raise if log_level is wrong type
* silly goose, loglevel is a subclass of str
* i don't believe this code
* add guard
---------
Co-authored-by: Khaleel Al-Adhami <khaleel.aladhami@gmail.com>
* make object var handle all mapping instead of just dict
* unbreak ci
* get it right pyright
* create generic variable for field
* add support for typeddict (to some degree)
* import from extensions
* Allow custom app module in rxconfig
* what was that pyscopg mess?
* fix another mess
* get this working with relative imports and hot reload
* typing to named tuple
* minor refactor
* revert redis knobs positions
* fix pyright except 1
* fix pyright hopefully
* use the resolved module path
* testing workflow
* move nba-proxy job to counter job
* just cast the type
* fix tests for python 3.9
* darglint
* CR Suggestions for #4556 (#4644)
* reload_dirs: search up from app_module for last directory containing __init__
* Change custom app_module to use an import string
* preserve sys.path entries added while loading rxconfig.py
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
* Show file and line number in deprecation warnings
* Exclude modules/packages by import
Less bespoke method of considering some packages to be part of the framework
and passed over when finding user code.
* Enable automatic retry on redis errors
ExponentialBackoff 3x retry for BusyLoadingError, ConnectionError, and TimeoutError
* retry on any redis error
* Use default single-retry for any RedisError
Using the default Retry means that async and sync clients get the appropriate type of Retry
* Proxy backend requests on '/' to the frontend
If the optional extra `proxy` is installed, then the backend can handle all
requests by proxy unrecognized routes to the frontend nextjs server.
* Update lock file
* pre-commit fu
* AppHarness: set config frontend_port and backend_port
* integration: frontend port and backend port should return the same content
with proxying enabled by default in dev mode, both frontend and backend ports
on / should return the same content.
* Retry up to 100 times when proxying to frontend
* Reduce retry attempts to 25
Fix log level passing to subprocess
* scripts/wait_for_listening_port: primarily check HTTP responses
if the port is up or not, we don't really care... the HTTP request needs to
work and not return errors
* aiohttp is an optional dep
* adapt integration.sh for --backend-only (counter integration test)
* woops
* windows WTF?
* scratching my head 🎄
* double WTF windows
* Fix remaining integration tests
* BaseState.get_var_value helper to get a value from a Var
When given a state Var or a LiteralVar, retrieve the actual value associated
with the Var.
For state Vars, the returned value is directly tied to the associated state and
can be modified.
Modifying LiteralVar values or ComputedVar values will have no useful effect.
* Use Var[VAR_TYPE] annotation to take advantage of generics
This requires rx.Field to pass typing where used.
* Add case where get_var_value gets something that's not a var
* Throw warnings when Redis lock is held for more than the allowed threshold
* initial tests
* fix tests and address comments
* fix tests fr, and use pydantic validators
* darglint fix
* increase lock expiration in tests to 2500
* remove print statement
---------
Co-authored-by: Khaleel Al-Adhami <khaleel.aladhami@gmail.com>
* Avoid double JSON encode/decode for socket.io
socket.io (python and js) already has a built in mechanism for JSON encoding
and decoding messages over the websocket. To use it, we pass a custom `json`
namespace which uses `format.json_dumps` (leveraging reflex serializers) to encode the
messages. This avoids sending a JSON-encoded string of JSON over the wire, and
reduces the number of serialization/deserialization passes over the message
data.
The side benefit is that debugging websocket messages in browser tools displays
the parsed JSON hierarchy and is much easier to work with.
* JSON5.parse in on_upload_progress handler responses
* add eradicate rules for commented out code
* remove output change
* fix pyi messed up indent
* fix pyi again
* fix layout docstring
* fix pyi_generator to remove commented out props from docs
* fix pyi_generator and regenerate some pyi
* fix double strip
* update all pyi
* try to fix stuff in pyi_gen
* whatever
* remove that maybe? i don't know
* fix that shit?
* fix more shit, idk
* better not see you ever again, extra line
* fix: handle default_factory in get_attribute_access_type, add tests for sqla dataclasses
* only test classes which have default_factory + add test for no default
