On py3.9 and py3.10, `dict[str, str]` and other typing forms are kinda
considered classes, but they still fail when doing `issubclass`, so
specifically exclude generic aliases before calling issubclass.
Fix#4424
Bonus fix: support upcasting of pydantic v1 and v2 models
* wip rx._x.asset improvements
* only add symlink if it doesn't already exist
* minor improvements, add more tests
* use deprecated Generator for python3.8 support
* improve docstring
* only allow explicit shared, only validate local assets if not backend_only
* fix darglint
* allow setting backend only env to false.
* use new is_backend_only in assets
* ruffing
* Move to `rx.asset`, retain old API in `rx._x.asset`
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
* [ENG-3953] Support pydantic BaseModel (v1 and v2) as state var
Provide serializers and mutable proxy tracking for pydantic models directly.
* conditionally define v2 serializer
Co-authored-by: Khaleel Al-Adhami <khaleel.aladhami@gmail.com>
* Add `MutableProxy._is_mutable_value` to avoid duplicate logic
* Conditionally import BaseModel to handle older pydantic v1 versions
* pre-commit fu
---------
Co-authored-by: Khaleel Al-Adhami <khaleel.aladhami@gmail.com>
* Rework Init workflow
* minor format
* refactor
* add comments
* fix pyright alongside some improvements
* add demolink for blank template
* fix darglint
* Add more templates and keep template name in kebab case
* revert getting other templates since we'll use the submodules approach
* remove debug statement
* Improvements based on standup comments
* Add redirect logic
* changes based on new flow
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
When an event/event spec is marked as "temporal", it will not be queued unless
the backend is up. This can be used to prevent periodic events (like from
`rx.moment`) from queueing up while the backend is down, and then stampeding
when the backend comes up and the queue is drained. It can be used to avoid
processing many periodic events at once when the app is only expecting to
process such an event every so often.
* Handle Var passed to `rx.toast`
If the user provides a `Var` for `message` then apply it as `props["title"]` to
avoid a var operations error.
* remove unnecessary parentheses
* remove weird hacks
* get it right pyright
---------
Co-authored-by: Khaleel Al-Adhami <khaleel.aladhami@gmail.com>
In some cases, a routing failure can cause the failure to be cached. When the
router has a cached failure, pushing such a route will never call
routeChangeComplete, and thus on_load event will never be fired for that route.
Purposely clearing the error from the router allows the page to properly load
on subsequent attempts without refreshing the app.
The `rich` module should be set to `None`, indicating that rich should not be used.
Setting it to `False` worked before, but recently added code in typer fails
when checking `if rich is not None`.
ref: https://github.com/fastapi/typer/pull/847
* add typing to function vars
* import ParamSpec from typing_extensions
* remove ellipsis as they are not supported in 3.9
* try importing everything from extensions
* special case 3.9
* don't use Any from extensions
* get typevar from extensions
Because of some dodgy logic in Base.get_value and State.dict / State.get_delta
when the value of some state var X happened to be the name of another var in
the state Y, then the value for X would be returned as the value of Y.
wat.
Fixes#4369
* Temporarily downpin @radix-ui/themes <3.1.5
A visual/style regression was introduced in @radix-ui/themes 3.1.5
as described in radix-ui/themes#627 which reflex needs to avoid.
* Get expected radix library version from component
These are useful for typing purposes and should be exposed at the top level to
avoid requiring deep imports from subpackages that we may need to change later.
* Codeblock cleanup in markdown
* Initial approach to getting this working with rx.memo and reflex web
* abstract the map var logic
* the tests are not valid + pyright fix
* darglint fix
* Add unit tests plus mix components
* pyi run
* rebase on main
* fix darglint
* testing different OS
* revert
* This should fix it. Right?
* Fix tests
* minor fn signature fix
* use ArgsFunctionOperation
* use destructured args and pass the tests
* fix remaining unit tests
* fix pyi files
* rebase on main
* move language regex on codeblock to markdown
* fix tests
---------
Co-authored-by: Khaleel Al-Adhami <khaleel.aladhami@gmail.com>
* allow for event handlers to ignore args
* use a constant
* dang it darglint
* forgor
* keep the tests but move them to valid place
* add metainfo to keyevent
* format code
* fix pyi files
* generate all prefixes of event types
* change format rule
* fix and test bug in config env loading
* streamline env var interpretation with @adhami3310
* improve error messages, fix invalid value for TELEMETRY_ENABLED
* just a small hint
* ruffing
* fix typo from review
* refactor - ruff broke the imports..
* cleanup imports
* more
* add internal and enum env var support
* ruff cleanup
* more global imports
* revert telemetry, it lives in rx.Config
* minor fixes/cleanup
* i missed some refs
* fix darglint
* reload config is internal
* fix EnvVar name
* add test for EnvVar + minor typing improvement
* bool tests
* was this broken?
* retain old behavior
* migrate APP_HARNESS_HEADLESS to new env var system
* migrate more APP_HARNESS env vars to new config system
* migrate SCREENSHOT_DIR to new env var system
* refactor EnvVar.get to be a method
* readd deleted functions and deprecate them
* improve EnvVar api, cleanup RELOAD_CONFIG question
* move is_prod_mode back to where it was
* improve object var symantics
* add case for serializers
* check against serializer with to = dict
* add tests
* fix typing issues
* remove default value
* older version of python doesn't have assert type
* add base to rx field cases
* get it from typing_extension
* Refactor initialize_gitignore to support list type for files_to_ignore and improve current ignore handling. Dont sort the gitignore file.
* more consistent list comprehension var
* raise error when get package manager is not found
* add escape hatch
* handle installing frontend packages more gracefully
* fix no return
* dang it darglint
* [ENG-1104] patch `json.dumps` to handle `__wrapped__` objects
Unwrap proxied objects if the default serializer doesn't work.
* pre-commit fixup
* Skip default fallback logic when `cls` is specified
`cls` will provide its own default serialization mechanism, passing a `cls`
Encoder class is now also a way to opt-out of our patching shenanigans and just
use your own code.
This will work, provided the library doing the JSON encoding isn't also using
their own custom class.
* Override JSONEncoder.default instead of json.dumps
Many libraries (like httpx, used by openai), will use `from json import dumps`,
and if they do that before `reflex.state` gets imported, then they will get the
original dumps function instead of our patched one.
To workaround this, monkeypatch the `JSONEncoder.default` function instead.
This is also nicer behavior for custom subclasses of JSONEncoder; if someone
wants to opt-out of our monkeypatching, they can simply not call
`super().default(o)` in their subclass, which by default only raises a
TypeError.
---------
Co-authored-by: Nikhil Rao <nikhil@reflex.dev>
* type check for event handler if spec arg are typed
* fix the typecheck logic
* rearrange logic pieces
* add try except
* add try except around compare
* change form and improve type checking
* print key instead
* dang it darglint
* change wording
* add basic test to cover it
* add a slightly more complicated test
* challenge it a bit by doing small capital list
* add multiple argspec
* fix slider event order
* i hate 3.9
* add note for UnionType
* move function to types
* add a test for type hint is subclass
* make on submit dict str any
* add testing for dict cases
* add check against any
* accept dict str str
* bruh i used i twice
* escape strings and print actual error message
* disable the error and print deprecation warning instead
* disable tests
* fix doc message
---------
Co-authored-by: Khaleel Al-Adhami <khaleel.aladhami@gmail.com>
* Shiki codeblock support decorations
* add decorations to useEffect
* fix pyright
* validate decorations dict
* Fix exception message plus unit tests
* possible test fix
* fix pyright
* possible tests fix
* cast decorations before creating codeblock
* `plain` is not a valid theme
* pyi fix
* address PR comment
* component as literal vars
* fix pyi
* use render
* fix pyi
* only render once
* add type ignore
* fix upload default value
* remove testcases if you don't pass them
* improve behavior
* fix render
* that's not how icon buttons work
* upgrade to next js 15 and remove babel and enable turbo
* upload is a silly guy
* woops
* how did this work before
* set env variable
* lower it even more
* lower it even more
* lower it even more
* only do literals as component vars
* add existing path subclass for env checks
* use the power of dataclasses
* use metaclass
* fake the class name
* use annotated
* use flag instead of dict
* dang it darglint
* cleanups
These props are NOT event triggers themselves, but rather they are props that
expect a list or dict of event handlers.
Additional fix for calling an `@rx.event` decorated event handler with no
arguments.
