The error recovery process is necessary, but is particularly subtle. An
error may occur basically any time some user code is called, which may
leave the internal owl state in an invalid situation.
In this commit, we try to improve Owl behaviour when a crash occurs in a
onMounted hook. Before this commit, it was possible for other component
mounted hooks to be called even though the component was not mounted, or
for willUnmount hooks to be called even though the component was
mounted, but its onMounted hooks had not been called.
All these situations need to be handled when we catch an error. In this
commit, we try to clean more internal state in the error handling code,
to make Owl more consistent.
If a component onMounted hook causes an error, and an error handler
tries to reinitiate a new correct render, the fiber may be recycled. But
the scheduler removes it from its own set of tasks, so it will be
ignored.
We simply check for this condition before removing it from the tasks
set.
In this commit, we extend the owl App class to support multiple sub
roots. This is useful for situations where we want to mount sub
components in non-managed DOM. This is exactly what the Knowledge app is
doing, with mounting views in an html editor.
Currently, this requires some difficult and fragile hacks, and still,
the result is that it is very easy to mix components from the main App
and a SubApp. But Knowledge does not actually care about creating a sub
app. It only needs the possibility to mount sub components in dynamic
places.
closes#1640
Before this commit, when willStart/willUpdateProps took more than
3s, a console.warn was done. In odoo, when a warning is logged
during a test, the test fails and the build is considered as "in
error".
There's a component that loads several resources (sequentially) in
its onWillStart, which *sometimes* takes more than 3s, making
builds fail non deterministically. Since a recent change (which
adds another call in the problematic onWillStart), the warning gets
logged quite often.
A quick fix is necessary, so we change the warn into a log, which
won't make build fail.
We may consider alternatives in the future though:
- add a parameter to onWillStart, to disable the timeout, or to
specify the delay (which is 3s by default)
- do not warn in test mode
Previously, if you wanted to pass a prop and have it be translated, you
had to either to the translation manually in JS, or use a workaround
with t-set and a body so that Owl would translate it for you, and then
pass the t-set variable as a prop. This is quite inconvenient and is a
common use case.
This commit introduces the `.translate` suffix to solve this issue. When
a prop uses this suffix, it is treated as a string instead of a JS
expression, avoiding the need for quotes as well as their escaping and
allowing extraction tools such as babel to generate a clean string as
the term's translation id. This is also more ergonomic. This suffix is
available for both component props and slot props.
This change will still require some work in Odoo to correctly extract
the terms for props using this suffix.
Previously, the async hook warning that's emitted when async hooks take
too long to resolve would be emitted even if the component was
destroyed or cancelled. We used to check if the node's fiber was still
the current fiber as a proxy for cancellation, but this is insufficient
in the case where an app might be destroyed.
This commit also wraps the content of some of the warning tests into a
try/finally block so that a failure of these tests doesn't pollute
setTimeout and console.warn and cause other tests to timeout one after
another.
Previously, we had a fixed whitelist for types that were allowed during
props validation. The implementation however supports using arbitrary
classes, and in practice it's desirable to do so, and already done when
not using typescript (when using typescript, it will error if the class
is not whitelisted), eg in Odoo, we use "Element" for the arch in the
standard view props, but this causes all view controllers to fail type
checking because it's not whitelisted.
This commit simply replaces existing constructors by a generic
constructor type, and adds a test with a validation success and a test
with a validation failure.
Before to this commit, the .trim modifier did not work correctly. The value
in the model is always trim at each input event but not the visual value
in the input. This difference between the visual and the model value cause
few strange bug.
After reflection, we think that we always want to be in .lazy when we use .trim.
Because we want to trim the final value during the onchange event and not
at each input event. If we do it at each input event, we can't write more then
one word easily.
So this commit change the behavior of the .trim modifier to always be in .lazy
Previously, having a Function as a type in the static props description
of a component would only work if the component was not the root
component, as the static props description on Component was "any",
whereas the static props description on ComponentConstructor was
"Schema". This meant that static props description on non-root
components was not type-checked, and on root components it was
type-checked only on the mount call.
This commit makes it so that static type description is of type "Schema"
on Component, now causing static props description to be type-checked,
and adds `typeof Function` to the `BaseType` union, which allows
declaring that a component expects a function as a prop.
Closes#1448
Previously, if a t-set-slot was inside a t-component itself inside a
t-set-slot the parser would crash, because the slot is removed from the
template before compiling its content, causing a further check's
assumption to be broken (the t-set-slot remains connected to the
component's xml node)
`useEffect` type is not compatible with later versions of Typescript.
(I tried with Typescript 5.2, I didn't check which specific
version had the breaking change)
This prevents other projects using owl (such as o-spreadsheet) to
upgrade their own version of Typescript.
`<T extends [...T]>` raises the two following errors:
`Type parameter 'T' has a circular constraint.ts(2313)`
`A rest element type must be an array type.`
Firefox will not serialize the xmlns attributes of node inside
XMLDocuments, see https://bugzilla.mozilla.org/show_bug.cgi?id=175946
it will only output an xmlns attribute if the node being serialized has
a non-null namespaceURI.
When compiling templates, we currently use a special attribute,
"block-ns" to keep track of the namespace, but we do not make use of the
namespace to create the elements that will be serialized to the compiled
block string. This causes a difference in behaviour between Chrome and
Firefox: since we end up with an Element with two attributes: the
block-ns attribute and the xmlns attribute. Chrome will serialize both,
but Firefox will only serialize the block-ns because the Element does
not have a namespaceURI.
To fix this issue, we get rid of the block-ns magic attribute
completely, and use xmlns instead. We also use the namespace when
creating intermediate elements that will be used to create the
serialized block string: the namespace is only used to create the
elements but *not* set as an attribute, as this would cause chrome to
serialize it twice, causing a duplicate attribute error when parsing it
further down the line.
When creating the template element for the block at runtime, the xmlns
attribute is used both as the namespace with which to create the
element, and set as an attribute, this doesn't cause issues when
serializing later because the namespaceURI is never serialized as an
attribute when serializing an HTML document[1], so we avoid the
double-serialization in Chrome, and when doing HTML serialization,
Firefox will correctly serialize the attribute.
It's desirable that the xmlns is set as an attribute to allow users to
use owl to render SVG, and then use the HTML serialization of it as a
SVG even outside the context of an HTML document (eg to generate an SVG
file or an SVG data URL).
[1]: https://w3c.github.io/DOM-Parsing/#xml-serialization
Because the default content of a t-call is compiled before the block of
the t-call is created, the default content can sometimes incorrectly be
treated as though it is at the root of the template. This causes various
issues, in the case of a t-if, the default content will just replace the
t-call entirely when truthy, and when falsy the template will not return
anything, leading to a crash. Similar things occur with t-foreach and
t-out with a default content.
This commit fixes that by moving the block creation for the t-call ahead
of the compilation of the default content. In doing so, it makes the
context attribute "preventRoot" obsolete: the purpose of this attribute
is to somehow make the code aware that the root of the template will be
defined later, but because it wasn't propagated everywhere properly it
caused this issue. Now that the root already exists before we compile
the default content, we don't need it anymore.
In odoo/owl#1352 we added support for using t-foreach on Map objects,
maps should behave the same as objects where keys are available under
the name specified in t-as and values under that same name with the
suffix `_value`. Unfortunately, because the code for objects was written
in a confusing way (values were stored in a variable named `keys` and
vice versa), the implementation for Map was incorrect and keys and
values were swapped.
This commit fixes that and rewrites the code to be less confusing: keys
are extracted from the variables `k_block` and values from `v_block`,
which swaps the behaviour, and the code to prepare a list from an object
now extracts the keys in `keys` and the values in `values`
Previously, when a template failed to compile because of a syntax error
(typically because we don't do any syntax checking when compiling
expression, allowing invalid expressions to be transpiled successfully),
the error reporting was very minimal: you would only get the error
message itself (eg: "Unexpected token") with the stack information of
the error pointing to the call to `new Function` in owl, which is not
very useful.
This commit catches the compilation error and completes it with
information about the template name when available, and also adds the
generated code to the error message, allowing the user to just
copy/paste it in their web console or code editor to get a more precise
location for the error.
Before this commit, most of the time, components are destroyed when the
virtual dom is patched and a component node is removed. However, since
Owl is asynchronous and a component may take some time to get ready
(with onWillStart), it can happen that a component is created, then
before it is ready, it is recreated. In that case, the initial instance
has to be destroyed.
Before this commit, the destroy operation was done immediately, when we
cancel the current fibers. However, this means that we cannot have a
guarantee between micro task ticks that a component has not been
destroyed in the meantime.
For example, in Odoo, it is common to use the rpc service, which will
throw an error if called by a destroyed component. But because of the
possible destruction of a component at any time, the following code is
unsafe:
async loadSomeData() {
// guaranteed to be called when component is alive
await Promise.resolve();
// however here, component may have been destroyed
this.rpc(...)
}
So, to prevent this issue, we can slightly delay the destroy operation.
It is not entirely trivial, since we need to find a way to neutralize
the component in the meantime. But it seems like performing all that
kind of operation at the "commit" phase (so, the request animation frame
callback) makes sense to me.
So, this commit modifies the code to add a new component status
(cancelled) and use it to cancel components that are waiting to be
destroyed. These components will then be destroyed as soon as the
requestanimation frame starts, before all other dom operations.
Before this commit, owl parser would ignore the `.number` suffix on
<select> options. I do not see a good reason for that, and it prevents
some legitimate usecases.
closes#1444
Before this commit, when using a dynamic t-call, it was possible to have
a collision between subcomponent keys. This would cause hard to
understand owl crash, because owl would incorrectly use the same
component instance for two different template location. From
owl point of view, one of these component has to be destroyed and the
other one should be mounted, so a crash would occur.
The fix is to simply properly key sub components.
Before this commit, in some specific situations, owl could skip updating
the DOM, even though it should. Here is what could happen:
- we have a parent component (A) and a child component (B)
- B depends on some reactive props from A
- we update some state that results in A being rendered, and B updated
- before this render is applied to the dom, we update some state again,
which causes A to be rendered again
- however, this new render is such that the B props are now identical,
so it will skip rendering B
- it will then apply the result of the render to the DOM => only A is
updated but B should also be updated!
The problem comes from the fact that Owl committed the props to the
component right after the rendering, to the new props were used in the
props comparison method to decide if B should be updated. This is
incorrect, we should always use the current props to decide what to do,
and only commit them to the component after it has been patched.
In JS, objects are often used as a mapping that gives fast lookup on the
keys, in those cases, the keys themselves are not known in advance, but
the values may have a shape that they're expected to follow. This commit
adds support for a `values` key in schema for objects, which will be
checked against all values in the object during validation.
It is not really useful, but it is possible to bind an anonymous
function prop. However, with the recent change on how the bind feature
works, it now crashes.
This commit makes sure that we properly apply the `bind` operation to
the function, and not to the last term of the function.
It is common in Owl components to have anonymous function as props.
However, since each rendering create a different (but equivalent)
closure, Owl will consider the props different, so will update the child
component, but this is (often) not necessary.
This commit will help reduce the problem by introducing a new `.alike`
prop suffix, that will let Owl know that each version of that specific
prop should be considered the same, so, will be ignored by the props
comparison code.
closes#1360
Previously, refs could get set to null incorrectly, or could stay set
when they shouldn't. This was caused by the fact that singleRefSetter
and multiRefSetters are created on every render, meaning that any render
that did not affect the status of the ref (mounted or not), would
overwrite the previous ref setter, including its closure, causing the
captured `_el` or `count` to be lost. This means that on a subsequent
render that did affect the status of the ref, the singleRefSetter would
consider that it did not set the ref, and so shouldn't unset it, causing
it to incorrectly stay keep the element, while in multiRefSetter, the
opposite problem occured: the count would be 0 on removal, causing it to
believe that it was the first call in the corresponding patch that
affected the ref, when in fact, the closure is already stale, because it
was created from the previous render, and the fresh multiRefSetter from
the latest render may already have been called by an element with that
ref that came earlier in the template.
This commit changes the ref setting strategy to work around the issue of
stale closures: we define a setRef method on ComponentNode that is
always called, this method ignores calls with `null`, meaning that the
refs object on the ComponentNode never reverts to a null value for any
key. Instead, the useRef hook will check whether the element that the
ref points to is still mounted, and return null if not.
Before this commit, properties were just handled as a special case of
attributes. But it does not make that much sense, since they are
different. For example, the `readOnly` property does not have the same
name as the `readonly` attribute. The confusion probably comes from the
fact that QWeb does not distinguish between property and attributes, so
Owl has to infer which one is which.
With this commit, we introduce a `block-property` directive in blockdom,
and change the compiler to use it in emitted code. It has the additional
benefits of slightly shrinking the runtime code, since the `isProp`
function can now be located in the compiler.
In #1161, we fixed blockdom treating 0 as falsy when patching a value,
which would result in an empty attribute. It turns out that there is
another place where we had the same fallback, which is when we compute a
dynamic attribute value, so while blockdom had the ability to set the
value to 0, when using a dynamic attribute value we would never give it
0 as a value. This commit changes the fallback strategy for dynamic
attribute values to be the same as the one in blockdom.
closes#1358
Have a t-ref on a DOM node with a t-key.
Change the t-key.
Before this commit, the old block removed its element from the component's refs *after*
the new block had been mounted, meaning that in effect, the resulting
ref at the end of the whole patch was null.
After this commit, we only remove an element from the component's ref if that very same
element was indeed the ref. (otherwise it means someone else has changed the ref.)
In dev mode, owl inserts an additional props validation step. Before
this commit, the validation would iterate on all key/value pairs of each
props. If the props is reactive, this has the unfortunate side effect of
subscribing the component to each of the keys, even though it should not
be.
This commit avoids the issue by only validating the raw object.
Previously, the getSubscriptions debugging utility returned all observed
keys for a given target, instead of only the keys observed by the
callback it received as argument. This commit fixes that issues.
Previously, if the value of a t-component directive changed, but there
was already a scheduled render for the component before that change, the
rendering of the existing component would get delayed (as it should),
but after the parent's rendering was complete, the old component which
will be destroyed during the patch would still get rendered, despite
being stale. This can cause crashes if that component relies on state
that no longer exists.
This was caused by the fact that when rendering, we check the parent
chain for an active render, and also check that the component still
exists in the parent's fiber childrenMap (ie, we know that the upcoming
patch is not about to destroy the component). To do this, we use the
component's parentKey, but in the case of t-component, the parentKey for
both possible components is the same, causing the stale component to
incorrectly believe that it's not about to be destroyed, by finding the
next component in the childrenMap under the same key.
This commit fixes that by prepending the component's name to the
parentKey, meaning that if the value of the t-component changes, so will
the key, and the render will be further delayed until the new state is
patched, at which point the stale component will be destroyed and the
following attempt to render will be cancelled as expected.
The choice to use the component's name is to simplify the
implementation, this means that if using t-component and switching
between components that have the same class name, this protection will
not work. The alternative would be to generate a unique id for the
component class, eg with a WeakMap, but this both complicates the
implementation and adds runtime overhead, for a case that is likely
extremely rare. We are open to refine the implementation should this
problem occur in practice.
Have a radio group defined inside a t-foreach:
```xml
<t t-foreach="values" t-as="val" t-key="val">
<input name="radiogroup" t-att-value="val" t-model="state.radioGroup" />
</t>
```
Before this commit the algorithm that set the "checked" attribute on the current active
radio button according to the state did not support having a dynamic value (`t-att-value`)
After this commit, this use case works as we go look in the dynamic attributes too.
t-call-context is a feature that's supposed to mask the rendering
context completely, but currently the component remains available
through `this`.
This commit stops treating `this` as a reserved word, so that it's
compiled to a lookup in the rendering context, and adds `this` to the
rendering context when binding the component's rendering function. With
these changes, `this` behaves the same as before when outside a
t-call-context, but when the rendering context is overriden, the
template can no longer access `this`. `this` still represents the
instance of the component inside of the rendering function since it's
needed by owl internally.
A side-effect of this change is that now the rendering context is no
longer the instance of the component by default, but is always an object
with the component in its prototype chain. This was already the case
before in some contexts (eg inside t-foreach, or inside components with
a t-set/t-call anywhere in its template). This can cause issues in rare
cases when a component method was called directly on the rendering
context, as before this change, the method's bound this would be the
component instance (except in a t-foreach, component with a
t-set/t-call, etc), while after this change it is now never the
component instance. When the method only reads on `this` there is no
issue as all the components properties are available on the rendering
contexts, but setting a value on `this` will write on the rendering
context and not the component which is likely a mistake.
While this is a breaking change, simply adding a t-set/t-call to any
template would break components that would be broken by this change,
with this in mind we decided to make this change anyway so that
developers get the error as early as possible in the development cycle
rather than having a seemingly inocuous change break code under them.
Previously, when using a component with a slot within a t-call with
t-call-context, the component would become available again inside the
slot despite the t-call-context. This was caused by the fact that the
capture helper function creates an object with the component as its
prototype which is incorrect. It should just use the previous context as
its prototype.
With this commit, we make sure that the code for t-model is run before
t-on-input event handler. This is useful to make sure that the state
reflected by t-model is up-to-date.
closes#1295
With the introduction of t-call-context, there is now no guarantee that
the ComponentNode can be found in the rendering context, any attempt to
do so can crash when combined with t-call-context. This commit fixes
that by using `this` instead, which in compiled templates refers to the
component that is being rendered.
When a parent component has an empty child with a t-on
and an event is triggered inside the parent, blockdom
checks if there is an event_catcher to call.
Doing so, an empty child would cause an error.
This commit fixes that.
Currently, the `moveBefore` method on VNodes assumes that the `other`
VNode it receives is of the same type, and that the entire VNode tree
below that other VNode has the exact same structure. While this is
correct in most cases, it breaks down when there is a VToggler somewhere
in the VNode tree, as the structure below a VToggler can be very
different from the structure below another VToggler that was created
from the same compiled code. For example, two iterations of a t-foreach
that contains a <t t-component="..."/> may spawn different components,
and different components obviously have different structures.
One way to fix this is to remove the assumption that the structure of
the `this` block tree in moveBefore is the same as the structure
of the `other` block tree, and instead, always give the concrete DOM
node before which we want to move the current VNode instead of giving it
a VNode and an afterNode as a fallback. One problem with this solution
is that it degrades performance in the "standard" case, where a
t-foreach contains no VToggler anywhere in its block tree, as retrieving
the first concrete DOM node requires calling firstNode() which
recursively traverses the entire tree.
To avoid this performance penalty in the standard case, we opt to only
go down this route whenever we encounter a VToggler when calling
`moveBefore`. This requires that we maintain two separate methods, one
to move a VNode before another VNode of assumed similar structure, which
is basically the current implementation of `moveBefore` for all VNode
types except VToggler, and one implementation that moves a VNode before
a concrete DOM node. This method needs to be implemented for all VNode
types, as all VNode types can be descendants of a VToggler. This method
will only be called from one place: the `moveBeforeVNode` method of the
toggler, which is the point where we realize that the assumption of
identical structure breaks down.
Co-authored-by: Bruno Boi <boi@odoo.com>
Currently, if a component has a default slot defined with t-set-slot,
and also content that compiles to something (eg, text or even a comment
node), the content takes priority over the t-set-slot. As t-set-slot is
more explicity, it should have priority.
Currently when checking for duplicate keys, we insert the value of the
key as is in a set then check for unicity against those. When the key is
an object, we check for duplicates based on object identity, whereas the
keys are used by owl as strings, and so using objects can cause
duplicate key errors that do not throw correctly but crash in the owl
internals.
This commit fixes that by making the duplicate checking code serialize
the key to string before insertion and when comparing against existing
keys.
The t-out directive is compiled internally into a LazyValue, which
represents a value that may or may not be created sometimes in the
future. It can also be reused more than once, and this is where there
may be an issue: if a component is contained in the lazyvalue, it needs
a unique key (coming from the t-foreach) to be properly indexed in the
parent children map. However, the LazyValue does not keep the key
information, so it is not able to provide it to its content.
The fix is then quite clear: the LazyValue class should store the key
information, and provides it to its content. This allows the LazyValue
to be used multiple times, in any place in a template.
closes#1270
A recent commit fixes the props validation code to make it work
regardless of the rendering context (important with the recent
t-call-context directive). Unfortunately, it then breaks props
validation through slots, because it assumed that the parent node in the
virtual node was the parent of the component, but it is not necessarily
true.
To fix this, we can use a simple property of the template functions:
they are bound to the current instance of the component, so we can
simply use "this"
The code for props validation assumed that the rendering context was a
component. This was actually true when it was written, but is no longer
true since t-call-context was introduced.
Because of that, it would crash when trying to access the internals of
the component, such as the static components object.
The fix is simple: instead of passing the context to the props
validation code, which can now be anything, we pass the component node,
which is guaranteed to give a reference to the component (and also to
the app). This also make the code slightly simpler.
closes#1261
This commit makes it so that when an error occurs in an owl app and none
of the registered error handlers are able to handle it, we rethrow the
error instead of just logging it to the console and swallowing it. This
allows users of owl to handle errors that happen in owl applications by
using event listeners for error and unhandledrejection events on the
window.
Previously, event handlers would not work when an app was mounted in an
iframe, this is caused by a guard in the event handler that checks that
the target element is still in the document, but it doesn't check
against the correct document in the case of an iframe.
This commit changes the check to check against the target's
ownerDocument.
Before this commit, a crash could occur when a component with no props
is defined in a slot, and that slot is conditionally displayed in
multiple locations.
The reason for that is that the key provided to the callSlot function
was identical, so from the perspective of the component function, it was
not possible to make the difference between a component located in
either places. With this commit, we make sure that a unique key is used
when a slot is reused in a template (or if it is dynamic, because in
that case, we have no idea at compile time if it will be unique or not)
closes#1246
Géry Debongnie