* feat: Optionally comparing fields, e.g. on rx.Base based types.
* feat: Minimally invasive change.
Leave the current implementation as is. Added test.
* fix: Supporting old-school python versions.
* fix: Adding masenf's suggestions to use var instead of string.
When setting the font_family prop for a component, also set the radix token
`--default-font-family` so that child radix components will inherit the font.
Always treat ComponentState subclasses as mixin, and explicitly pass
`mixin=False` in `.create` classmethod when intentionally creating a substate.
This allows a "base" ComponentState to have subclasses that also work as
ComponentState themselves.
Fix#3368
* add support for lifespan tasks
* allow passing args to lifespan task
* add message to the cancel call
* allow asynccontextmanager as lifespan tasks
* Fix integration.utils.SessionStorage
Previously the SessionStorage util was just looking in localStorage, but the
tests didn't catch it because they were asserting the token was not None,
rather than asserting it was truthy.
Fixed here, because I'm using this structure in the new lifespan test.
* If the lifespan task or context takes "app" parameter, pass the FastAPI instance.
* test_lifespan: end to end test for register_lifespan_task
* In py3.8, Task.cancel takes no args
* test_lifespan: use polling to make the test more robust
Fix CI failure
* Do not allow task_args for better composability
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
* wip connection toaster
* never duplicate toast for websocket-error
* wip update banner
* clean up PR
* fix for 3.8
* update pyi
* ConnectionToaster tweaks
* Use `has_too_many_connection_errors` to avoid showing the banner immediately
* Increase toast duration to avoid frequent, distracting flashing of the toast
* Automatically dismiss the toast when the connection comes back up
* Include `close_button` for user to dismiss the toast
* If the user dismisses the toast, do not show it again until the connection comes back and drops again
* Use `connection_error` var instead of a custom util_hook to get the message
* ConnectionPulser: hide behind toast
* Hide the connection pulser behind the toast (33x33)
* Add a title (tooltip) that shows the connection error
* Re-add connection pulser to default overlay_component
If the user dismisses the toast, we still want to indicate that the backend is
actually down.
* Fix pre-commit issue from main
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
* Support replacing route on redirect
Support next/router `.replace` interface to change page without creating a
history entry.
* test_event: include test cases for new "replace" kwarg
New experimental feature to create client-side react state vars, save them in
the global `refs` object and access them in frontend rendering/event triggers
as well on the backend via call_script.
* [REF-2787] add_hooks supports Var-wrapped hooks
* Fix VarData definition in .pyi file to allow removal of type ignore comments
* Var.create and Var.create_safe accept _var_data parameter
* Replace instances where a set of imports was being passed to VarData
* Update code throughout reduce use of `._replace` to add VarData
* Fixup: user hooks _var_data.imports will never be iterable, just a single ImportDict
* typed mixins
* implicit mixin=True kwarg for ComponentState subclasses
* fix: always init other subclasses
* adjust tests: all mixins support base vars now
* Get `action` and `cancel` working for rx.toast
Respect defaults set in ToastProvider toast_options when firing a toast with
it's own ToastProps set.
* Update reflex/components/sonner/toast.py
Co-authored-by: Thomas Brandého <thomas.brandeho@gmail.com>
* Move queueEvent formatting into rx.utils.format module
Implement on_auto_close and on_dismiss callbacks inside ToastProps
* Update rx.call_script to use new format.format_queue_events
Replace duplicate logic in rx.call_script for handling the callback function.
* Move PropsBase to reflex.components.props
This base class will be exposed via rx._x.PropsBase and can be shared by other
wrapped components that need to pass a JS object full of extra props.
---------
Co-authored-by: Thomas Brandého <thomas.brandeho@gmail.com>
* Throw Warning for Projects Created in OneDrive on Windows
* precommit
* remove dead code
* REFLEX_USE_NPM escape hatch to opt out of bun
In some unsupported environments, we need to just not use bun. Further
investigation needed.
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
The demo template includes a drop down menu item named Contact
has an email address to the founders. However, the email address
includes an extraneous = character in the domain part making it
invalid.
Signed-off-by: Eric Brown <eric_wade_brown@yahoo.com>
* Allow `Component.add_style` to return a regular dict
It's more convenient to allow returning a regular dict without having to import
and wrap the value in `rx.style.Style`.
If the dict contains any Var or encoded VarData f-strings, these will be picked
up when the plain dicts are passed to Style.update().
Because Style.update already merges VarData, there is no reason to explicitly
merge it again in this function; this change keeps the merging logic inside the
Style class.
* Test for Style.update with existing Style with _var_data and kwargs
Should retain the _var_data from the original Style instance
* style: Avoid losing VarData in Style.update
If a Style class with _var_data is passed to `Style.update` along with kwargs,
then the _var_data was lost in the double-splat dictionary expansion.
Instead, only apply the kwargs to an existing or new Style instance to retain
_var_data and properly convert values.
* add_style return annotation is Dict[str, Any]
* nit: use lowercase dict in annotation
* rx.accordion customizability overhaul
* fix low contrast font in "classic" variant
* relative font sizing, so setting font_size on the root works now
* use CSS vars wherever possible to make downstream styling easier
* optional dividers between accordion items
* support `radius`
* support `duration` and `easing` for controlling the animation
* fix animation jank while keeping padding_y
* lookup `data-variant` via CSS instead of passing props in python
* fix "surface" variant to use `--accent-surface` as radix intended
* Restore default variant: "classic"
* Fix accordion primitive radius
* Unit tests for add_style and component styles with foreach
The styles should be correctly applied for components that are rendered as part
of a foreach.
* [REF-2802] Foreach should respect modifications to children
Components are mutable, and there is logic that depends on walking through the
component tree and making modifications to components along the way. These
modifications _must_ be respected by foreach for consistency.
Modifications necessary to fix the bug:
* Change the hash function in `_render` to get a hash over the render_fn's
`__code__` object. This way we get a stable hash without having to call the
render function with bogus values.
* Call the render function once during `create` and save the result as a child
of the Foreach component (tree walks will modify this instance).
* Directly render the original (and possibly modified) child component instead
of calling the render_fn again and creating a new component instance at
render time.
Additional changes because they're nice:
* Deprecate passing `**props` to `rx.foreach`. No one should have been
doing this anyway, because it just does not work in any reasonable way.
* Raise `ForeachVarError` when the iterable type is Any
* Raise `ForeachRenderError` when the render function does not take 1 or 2 args.
* Link to the foreach component docs when either of those errors are hit.
* Change the `iterable` arg in `create` to accept `Var[Iterable] | Iterable`
for better typing support (and remove some type: ignore comments)
* Simplify `_render` and `render` methods -- remove unused and potentially
confusing code.
* Fixup: `to_bytes` requires `byteorder` arg before py3.11
* [REF-2789] Graceful deprecation of rx.input.root and rx.input.input
Handle previously valid code where rx.input.root wrapped
rx.input/rx.input.input and rx.input.slot.
Raise deprecation warnings with hints about how to refactor code.
Copy props from rx.input.root to children inputs and apply any rx.input.slot
components to children inputs in an attempt to keep existing code working as
best as possible.
Fix DebounceInput:
* pass children through (for rx.input.slots)
* pass _rename_props through (for color_scheme)
* Fix for case where `rx.input.root` had event triggers
Fix for case where `rx.input.root` had no input children
* radix themes progress: expose `max` prop
It doesn't work yet, but PR filed for radix-ui/themes:
https://github.com/radix-ui/themes/pull/492
* Move `progress` and `toast` to `_x`
[REF-2779] Expose skeleton and data_list in top level namespace.
* update blank template
* update CONTRIBUTING.md
* adjust min_height of main stack
* simplify stuff
* fix import order for blank template and tell ruff to ignore it
* upgrade to latest pip for in_docker_test_script.sh
* Bump gunicorn to 22.0.0 (security)
Changelog: https://docs.gunicorn.org/en/stable/news.html#id1
use utime to notify workers liveness
migrate setup to pyproject.toml
fix numerous security vulnerabilities in HTTP parser (closing some request smuggling vectors)
parsing additional requests is no longer attempted past unsupported request framing
on HTTP versions < 1.1 support for chunked transfer is refused (only used in exploits)
requests conflicting configured or passed SCRIPT_NAME now produce a verbose error
Trailer fields are no longer inspected for headers indicating secure scheme
support Python 3.12
** Breaking changes **
minimum version is Python 3.7
the limitations on valid characters in the HTTP method have been bounded to Internet Standards
requests specifying unsupported transfer coding (order) are refused by default (rare)
HTTP methods are no longer casefolded by default (IANA method registry contains none affected)
HTTP methods containing the number sign (#) are no longer accepted by default (rare)
HTTP versions < 1.0 or >= 2.0 are no longer accepted by default (rare, only HTTP/1.1 is supported)
HTTP versions consisting of multiple digits or containing a prefix/suffix are no longer accepted
HTTP header field names Gunicorn cannot safely map to variables are silently dropped, as in other software
HTTP headers with empty field name are refused by default (no legitimate use cases, used in exploits)
requests with both Transfer-Encoding and Content-Length are refused by default (such a message might indicate an attempt to perform request smuggling)
empty transfer codings are no longer permitted (reportedly seen with really old & broken proxies)
** SECURITY **
fix CVE-2024-1135
* Remove TYPE_CHECKING guard for pydantic v1 imports
Retain TYPE_CHECKING guard in v1 fallback to force pyright into pydantic.v1 namespace
* Run unit tests with pydantic v1 now that v2 is installed via poetry
* Allow EventHandler args to be partially applied
When an EventHandler is called with an incomplete set of args it creates a
partial EventSpec. This change allows Component._create_event_chain to apply
remaining args from an args_spec to an existing EventSpec to make it
functional.
Instead of requiring the use of `lambda` functions to pass arguments to an
EventHandler, they can now be passed directly and any remaining args defined in
the event trigger will be applied after those.
* [REF-2273] Implement `.setvar` special EventHandler
All State subclasses will now have a special `setvar` EventHandler which
appears in the autocomplete drop down, passes static analysis, and canbe used
to set State Vars in response to event triggers.
Before:
rx.input(value=State.a, on_change=State.set_a)
After:
rx.input(value=State.a, on_change=State.setvar("a"))
This reduces the "magic" because `setvar` is statically defined on all State
subclasses.
* Catch invalid Var names and types at compile time
* Add test cases for State.setvar
* Use a proper redis-compatible token
* test_foreach: assert on arg _var_type
* [REF-2682] Foreach over dict uses Tuple arg value
When iterating over a Var with _var_type dict, the resulting arg value
_var_type should be Tuple[key, value] so it can be correctly used with other
var operations.
Fix#3157
* Correct _var_type for iteration over Tuple of multiple types
The arg value when iterating over a tuple could be any of the possible values
mentioned in the annotation.
When only one type is used, the Union collapses to the base type, at least in py3.11
* Add comments
* pyi_generator: ignore `app.py` and single files
When running pyi_generator in pre-commit, it passes individual changed files on
the command line as targets, and these were not being properly excluded
according to the EXCLUDED_FILES list.
Add app.py to the EXCLUDED_FILES list so it does not get regenerated
automatically.
* [REF-2587] Ignore top-level theme appearance
From the Radix docs, it is not recommended to actually set `appearance`, but
instead use next-themes to set and switch the appearance dynamically.
Because Reflex already compiles the top-level theme into the next-themes
ThemeProvider, we can blank out the appearance prop after compiling contexts.js
to avoid a mismatch between the selected app appearance and the appearance in
the rx.theme when displaying overlay components.
Fix#2992
* Simplify `run_process_with_fallback` logic
Include log streaming for initial command and fallback command. This is
necessary because if the command produces significant output that is not
consumed, it can fill the OS pipe and block the process from running.
* Avoid overwriting next.config.js when content has not changed
Windows cannot seem to detect when the file changes but the
content is the same and triggers a much longer hot-reload
cycle to deal with the "updated next.config.js"
* Handle case where `next.config.js` doesn't exist yet
* add more tests
* add tests to raise coverage
* more tests, bump coverage to 73
* fix up icon_button test
* fix darglint for app.py
* fix utcnow usage warning
* set threshold to 72
* fix timestamp
* fix unit tests for linux-redis
* removed commented code and put a TODO
* adjust share command prompt
* use lower case "y" for console.ask, keep order consistent as "y" then "n"
* share command update: do not suggest demo URL is mandatory, also move to the last question; when checking if user has permission to upsert package, do POST to record this package (if first time sharing)
* clean up prompt for preview image
* build pyi files when building/publishing 3rd party
* fix typo in workflow
* add future annotation
* add tests to pass coverage check
* add more unit tests
* omit pyi_generator from test coverage
* change black from dev deps to direct deps
* remake all pyi
* format pyi if black is present, return as if otherwise
* fix requested changes
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
* Add test_get_state_from_sibling_not_cached
A better unit test to catch issues with refetching parent states
and calculating the wrong parent state names to fetch.
* _determine_missing_parent_states: correctly generate state names
Prepend only the previous state name to the current relative_parent_state_name
instead of joining all of the previous state names together.
* [REF-2219] Avoid refetching states that are already cached
The already cached states may have unsaved changes which can be wiped out if
they are refetched from redis in the middle of handling an event.
If the root state already knows about one of the potentially missing states,
then use the instance that is already cached.
Fix#2851
* Connection pulser only depends on has_connection_errors
Avoid showing the WiFi error icon when the state is hydrating / navigating
because not being hydrated is not indicative of a connection error in itself.
* Set is_hydrated=False at route onChangeStart
When navigation event starts, set is_hydrated=False on the client side before
any on_load event is dispatched. This avoids a flickering problem where the
client browser navigates and briefly shows content on the page before
processing on_load events associated with the page.
Fix#2885
* Update pyi
* Remove upper bounds of most dependencies.
Also adds a import try except block for pydantic.v1 and relocks.
Keep black and ruff to not mess to much with current formatting
Make pyright see the right import as long as constraint still lock pydantiv v1
Down pin pytest-asyncio again due to known issue
Fix upload handler with latest versions of fastapi
Change comment
* Add changed lockfile
* Set max versions for deps
* Revert app.pyi
---------
Co-authored-by: Malte Klemm <malte.klemm@blueyonder.com>
Co-authored-by: Nikhil Rao <nikhil@reflex.dev>
* [REF-2265] ComponentState: scaffold for copying State per Component instance
Define a base ComponentState which can be used to easily create copies of the
given State definition (Vars and EventHandlers) that are tied to a particular
instance of a Component (returned by get_component)
* Define `State` field on `Component` for typing compatibility.
This is an Optional field of Type[State] and is populated by ComponentState.
* Add integration/test_component_state.py
Create two independent counters and increment them separately
* Add unit test for ComponentState
* textarea: expose auto_height and enter_key_submit props
These two props improve the workflow for chat apps and other situations where
we want multiline input.
auto_height: resize the textarea based on its contents
enter_key_submit: pressing enter submits the enclosing form (shift+enter
inserts new lines)
Fix#1231
* Update pyi
* add command to share custom component info for gallery
* cleanup
* tested
* Update custom_components.py
Co-authored-by: Masen Furer <m_github@0x26.net>
* reorder the questions
* ask user to verify info before publishing
* remove questions already captured in pyproject.toml
* no need to cache form inputs since not many questions
* tested
* cleanup
* delete test
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
* parse_args_spec: resolve annotations with typing.get_type_hints
Ensure that event trigger specs in modules with future annotations are resolved
correctly.
Fix#2848
* Fix arg spec for on_upload_progress event
Make py3.8 compatible with `get_type_hints`
In exec.py, run_process_and_launch_url already passes `shell=True` on windows,
but the AppHarness has never really been used on Windows, so we were missing
this detail.
Fix#2941
* When a Var points to a model, prefer access to model fields.
When a Var points to a model, and fields of the model share the same name as
Var operations, access to the model fields is now preferred to avoid having Var
operation names shadow model fields. Since most Var operations do not actually
work against Models, this does not really block any functionality.
* Special case for ComputedVar needing to internally access fget
Since fget is a "slot" on property, normal __getattribute__ access cannot find it.
* Workaround https://github.com/python/cpython/issues/88459
In python 3.9 and 3.10, the `isinstance(list[...], type)` returns True, but
it's not a valid class for use in issubclass
* Use dill instead of cloudpickle for serialization
* smaller size pickles
* support dynamically defined states
* avoid issues with unpickleable globals
* pyproject: add dill, remove cloudpickle
* poetry.lock: relock dependencies
* Dynamically convert EventHandler to functools.partial
Instead of converting the functions up front and assigning them to the
instance, unbox the function from the EventHandler when it is requested via
__getattribute__. This reduces the size of the per-instance pickle, because
event handler bodies do not need to be included.
* Improve checking for cython_function_or_method
Because pydantic can be installed without cython, only use the workaround in
the case where the BaseModel.validate function is NOT a FunctionType,
indicating it's a cython function.
* Serialize all State subclasses by reference
* When __REFLEX_SKIP_COMPILE == "yes" allow telemetry to fail
Allow running `--backend-only` without a .web directory
* When re-initializing a pre-project_hash project, do not blow up on KeyError
* Revert "Remove width prop from rx.select (#2835)"
This reverts commit d14292dc9b.
* HighLevelSelect: pass flex_shrink prop to SelectTrigger
This allows the high level select to have width="100%" and also shrink-to-fit inside a parent container.
Fix#2824, kind of
* Revert "Revert "Revert "Revert "use process pool to compile faster (#2377)" (#2434)" (#2497)" (#2595)"
This reverts commit 6b6eea4d7d.
* Adjust number of operations for more correct progress bar
* app: recognize REFLEX_COMPILE_PROCESSES and REFLEX_COMPILE_THREADS
Control whether multiprocessing is used and the number of processes or threads
that should be used.
This will allow users to opt-in to the new, potentially hazardous,
multiprocessing mode, which results in much faster compiles, but has already
been reverted 4 times. Lets leave the code in this time, but use the thread
pool executor by default.
Limiting the number of threads or processes to 1 can also aid in debugging
issues that arise during compile time.
* Allow REFLEX_COMPILE_PROCESSES=0 to trigger multiprocessing with auto workers
* Separate `get_hooks` and `get_hooks_internal` for stable output
When downstream component wrappers depend on State when writing hooks, they
need to be assured that all internal hooks (events, var hooks, memoized
handlers, etc) will be rendered prior to user-defined hooks.
This also makes it less likely for downstream components to feel the need to
overwrite `get_hooks` (no underscore) directly and break internal functioning
of Reflex components.
* Include internal hooks in AppWrap and Page
* Apply get_hooks_internal in a few more places
Display a warning message, recommending python 3.11 with the current release of
reflex.
We need this until running with a uvicorn version that has better support for
py3.12 on windows.
* CustomComponent ignores the annotation type in favor of the passed value
Do not require rx.memo wrapped functions to have 100% correct annotations.
* test_custom_component_get_imports: test imports from wrapped custom components
* Account for imports of custom components for frontend installation
Create new ImportVar that only install the package to avoid rendering the
imports in the page that contains the custom component itself.
* Handle Imports that should never be installed
* Fix up double negative logic
* Perf Optimization: use the imports we already calculate during compile
Instead of augmenting _get_imports with a weird, slow, recursive crawl and
dictionary reconstruction, just use the imports that we compile into the
components.js file to install frontend packages needed by the custom
components.
Same effect, but adds essentially zero overhead to the compilation.
* Get `client_ip` from `asgi.scope`
It seems like REMOTE_ADDR is always 127.0.0.1, which is not super useful when
trying to figure out where the websocket connection is originating from.
Of course this isn't a silver bullet because most-likely the WS will be passed
through a reverse proxy anyway... in that case, the client IP is likely in the
headers under `x_forwarded_for`
* client_ip: fallback to REMOTE_ADDR
* Make @rx.memo work with state vars passed as props
Seems like this was a regression from the StatefulComponent refactor, because
trying to pass a state Var to a CustomComponent gave undefined, likely due to
`_get_vars` not accounting for `self.props` in CustomComponents.
With this change, it works.
Integration test added to `test_var_operations.py`
* Allow CustomComponent props to be Component
Avoid calling `.json()` on all Base types because the Var serializer already
does that, but this way, more specific types (like Component) can be serialized
differently.
When the type is Component, attach a VarData with the imports and hooks to when
the Var is rendered, it also carries the correct imports/hooks and does not
throw frontend errors.
* [REF-2158] Enable state when `on_load` or event_triggers are set.
Basically all events in a reflex app require the backend to be up, even the
built-in server side events round trip to the backend and require "state" even
if no user-defined state classes are declared.
test_app_state_determination: checking that state is enabled at the right time
* Clear out DECORATED_PAGES when initializing a new app
* rebase
* pass include_children kwarg in radix FormRoot
* respect include_children
* ruff fixes
* readd statemanager init, run pyi gen
* minor performance imporovements, fix for state changes
* fix pyi and pyright
* pass include_children for chakra
* remove old state detection
* add test for unused states in stateless app
---------
Co-authored-by: Masen Furer <m_github@0x26.net>
* [REF-1368] Move common form functionality to rx.el.forms
Allow plain HTML Form element to have magic on_submit event handler.
* Chakra and Radix forms inherit `on_submit` functionality from rx.el.form
Consolidate logic in the basic HTML form and use it in both Radix and Chakra
form wrappers.
* from __future__ import annotations for py38
* [REF-2098] Allow overlay Close components to have on_click handlers
The child of the Radix *Close components are passed `asChild`, meaning they are
treated as the direct child of the Close itself. This causes the `on_click`
handler attached to such an element to override the default `on_click` behavior
of the Close, which is to close the overlay, thus breaking the desired behavior
of the Close component.
When creating a Close component, if the child has an `on_click` handler, then
internally wrap it in a `Flex` so that the `Flex` gets the `asChild` treatment,
and both the default on_click and the child's on_click are both fired.
Bonus fix: pass `asChild` when constructing the `DrawerClose` to avoid a
<button> in <button> warning in the browser console, because the Drawer does
NOT pass `asChild` to the Close by default.
* Move Trigger logic to RadixThemesTriggerComponent base class
Apply trigger logic to Trigger, Close, Cancel, and Accept components; including
hover_card and dropdown_menu.
The ThemePanel fights with the ThemeProvider when the user color preference key differs from the `appearance` prop specified in the theme. To avoid issues when using the ThemePanel (in development), clear out the user color preference before loading the page and before unloading the page (to ensure it does not freeze on reload).
Clearing the user preference isn't ideal production behavior, but typically the
ThemePanel is only used during development for trying out different styles, and
having it not freeze the app is better dev behavior.
Fix#2650
