In many JS runtimes all closures created in the same scope share a context
this means that data held in one of the closures is not collected until all of the closures are collected.
This change prevents the returned promise from holding a reaction that holds the entire `Router` object in memory.
(cherry picked from commit 3867cd8554)
Deeply nested parentheses in URLs (e.g. `(a/(b/(c...)))`) trigger recursive calls in `UrlParser`, which can lead to a `RangeError: Maximum call stack size exceeded`. While such errors are generally caught by the framework, relying on the runtime's stack limit is unpredictable across different environments and engine states (e.g. varying stack sizes in different browsers or Node.js versions).
The deeply nested parentheses can cause a stack overflow. While such URLs can be valid (e.g., `(a/(b/(c...)))`) and serialize to simple paths (e.g., /a/b/c), excessive nesting is unreasonable and likely malicious or accidental.
Linear paths (e.g. /a/b/c/d) are parsed iteratively and do NOT trigger recursion. Only parentheses trigger recursion.
This commit introduces a recursion depth limit of 50. If parsing exceeds this depth, the router will now throw a specific `UNPARSABLE_URL` error with the message "URL is too deep". This ensures a deterministic failure mode that is easier for applications to handle than a crash or generic RangeError.
The limit of 50 is chosen as it should accommodate any reasonable application URL structure (including complex named outlets) while providing a safe upper bound against abusive payloads.
This is essentially a refactor of the error state:
* Before: RangeError (System says "I'm out of stack memory")
* After: RuntimeError (Validator says "Input is invalid")
This provides:
* Semantic Correctness: The error now correctly blames the input ("URL too deep"), not the environment ("Stack full").
* Cross-Platform Consistency: The limit is the same in Chrome, Firefox, Node, and Deno, regardless of their internal recursion limits.
* Fast Failure: We stop at depth 50 instead of depth ~15,000, saving those cycles (though CPU cost is negligible either way).
"wide" URLs are now theoretically more expensive than "deep" URLs (because deep ones fail fast), but both are well within safe bounds for any reasonable input size.
(cherry picked from commit 458bc4a2c8)
This commit prevents the Router from intercepting reload navigations
in the navigate event listener. This would convert hard page reloads
to SPA navigations.
fixes#66746
This commit introduces a highly requested `trailingSlash` configuration option to the Angular Router, allowing developers to control how trailing slashes are handled in their applications. The options are:
- 'always': Enforces a trailing slash on all URLs.
- 'never': Removes trailing slashes from all URLs (default).
- 'preserve': Respects the presence or absence of a trailing slash as defined in the UrlTree.
This publishes the work that was done to integrate with the Navigation
API as an experimental router feature. Browser support is limited and in
active development. There are also known bugs in the browser implementations
and only Chromium browsers supported deferred URL updates with the
`precommitHandler`. Relates to #53321, which I would likely not mark as
completed until this is at least in dev preview, which likely won't
happen until it is widely available and potentially delayed until
`precommitHandler` is widely available as well.
The final form of this api might not even be a "router feature" in the end, but instead be
something similar to what other frameworks have to provide different
platform integrations (e.g. `provideNavigationRouter`). That would
support omitting the history-based integration from the bundle when only
the navigation integration is used. Alternatively, the current
`provideRouter` could require one of `withHistory` or `withPlatformNavigation`.
Firefox-specific workaround: Firefox 144+ creates the `updateCallbackDone` promise
internally during error handling (e.g., duplicate view-transition-name), even before
the property is accessed. This can cause unhandled promise rejections to appear in
error tracking tools like Rollbar/Sentry. We must attach catch handlers to all three
promises immediately to prevent these unhandled rejections.
See: https://bugzilla.mozilla.org/show_bug.cgi?id=1999336 (if applicable)
Spec: https://drafts.csswg.org/css-view-transitions-1/#dom-viewtransition-updatecallbackdone
There was a bug introduced in #60875. While RouterLink doesn't
necessarily depend on Router state, its link can change when navigations
cause the `ActivatedRoute` paths to change. It is difficult to determine
which segments the link depends on. Luckily, the commit had flawed
logic:
```
!this.queryParamsHandling && !dependsOnRouterState(this.options?.defaultQueryParamsHandling);
```
The subscription gets created whenever `queryParamsHandling` was not
defined, or it was non-default. This is pretty much all scenarios since
nobody is likely to set it explicitly to the default value. In addition,
the `click` handler recomputes the tree because `urlTree` is a getter
that does the computation.
This commit effectively rolls back #60875
This change deprecates `Router.isActive` in favor of a function that
creates a computed which tracks whether a given url/UrlTree is active,
following changes in the router state.
`Router.isActive` contributes ~1333b to the bundle size, but is only
used by developers who use the API directly or who use
`RouterLinkActive`. It should not contribute to bundle sizes of
applications that do not use this functionality.
This builds off of #66197 by retaining the original navigateEvent across redirects
so the NavigateEvent can more accurately track the lifecycle of a navigation,
which may span across several NavigationStart events due to redirects
The `precommitHandler` of the Navigation API unlocks some of the truly
powerful features for Routers like Angular's which defer the URL
updates. Without the `precommitHandler`, we cannot initiate a navigation
until we are ready to commit the URL because it causes the URL to update
immediately.
With `precommitHandler` support, we are able to create a `NavigateEvent`
_immediately_ on navigation, which allows the browser to show that a
navigation is happening with a loading indicator. Site visitors will
also have the ability to cancel the navigation with the "stop" button.
When we are ready to commit the URL, the precommitHandler supports a
"redirect" function that we can use to first redirect the navigation to
a new location immediately before committing it.
The commit operation is not synchronous because the API waits for all
precommitHandlers to resolve. This commit adds a small bit of handling
to account for this so that the Router's transition does not advance
to the next stage until the URL has been committed.
Mutating the `UrlTree` that is returned by `createUrlTree` can cause the
input, which might even be an active route, to be mutated. This ensures
the `UrlSegment`s are recreated and do not mutate the input.
fixes#54624
This commit introduces a new feature to automatically destroy `EnvironmentInjector`s associated with routes that are no longer active or stored. This helps in managing memory by releasing resources held by unused injectors.
This change replaces all remaining occurrences of `typeof ngDevMode !== undefined`
with the correct `typeof ngDevMode !== 'undefined'` form. This aligns the codebase
with JavaScript typeof semantics and maintains consistency with other Angular code.
There is a bug (?) in all browsers where the timing of the entry change
is delayed when a navigation is initiated by a click on a link via user
interaction. In this case, we need to ensure we do a 'push' navigation
rather than a 'replace'.
Programatically doing element.click() does not reproduce
this behavior, so adding a test for this is difficult (would require
webdriver).
This commit adds a `.catch()` handler to `transition.finished` from `document.startViewTransition` to prevent unhandled promise rejections. The finished promise can reject with `TimeoutError` or `InvalidStateError` when transitions fail during or after the animation phase.
Based on the Blink source code, the `finished` promise can reject with:
* `TimeoutError`: "Transition was aborted because of timeout in DOM update"
* `InvalidStateError`: "Transition was aborted because of invalid state"
This may happen when the DOM update phase exceeds the browser's internal timeout threshold.
This adds additional tests following the initial implementation in #64737 that cover
the use-case of path parameters before and after the wildcard segment.
Unnecessarily aborting traversal navigations will break focus and scroll
restoration. We need to keep them open rather than replacing them with
an identical router-initiated navigation.
The end of the Router navigation is a block of synchronous logic that
can be compressed into a single operator rather than splitting it across
several, making it harder to step through. The only benefit from the
split is automatic unsubscribe/cancellation, which we can replicate
with an additional 'shouldContinue' check before proceeding.
Refactor to use async/await for clearer asynchronous operations and enhanced error handling.
Simplify resource caching and streamline the resolution of component templates and styles.
Update in the router to align with the new async resource resolution.
Wraps InjectionToken names with `typeof ngDevMode !== 'undefined' && ngDevMode`
checks to enable tree-shaking of descriptive token names in production builds.
This ensures debug-only strings are removed from production bundles, reducing size.
Also removes unused imports found during refactor.
Annotate the `new Version(...)` call with `/* @__PURE__ */` to signal to optimizers that the constructor is side-effect free.
Without this hint, bundlers such as Terser or ESBuild may conservatively retain the `VERSION` instantiation even when unused. With the annotation, the constant can be tree-shaken away in production builds if not referenced, reducing bundle size.
Replace `@Optional() link: RouterLink` constructor parameter with
`link = inject(RouterLink, {optional: true})` to enable tree-shaking
of the `Optional` decorator and its factory scaffolding.
Bundle size reduction: `Optional` is a runtime value created by
`makeParamDecorator()`. Even in production builds, ESBuild and other
bundlers must keep their factory code because it is referenced via
`Optional`. With `inject()`, this class is no longer referenced,
allowing it and the `makeParamDecorator` scaffolding to be tree-shaken
when unused elsewhere.
Note: This updates the constructor signature but should not be
considered a breaking change. Angular's official guidance is that
directives, components, and pipes should be instantiated by the
framework, not by user code. Directly calling `new RouterLinkActive(...)`
is an unsupported pattern that goes against Angular's design principles.
This adds handling of the abort event on the signal of the
`NavigateEvent`, allowing us to cancel the Router's ongoing navigation
transition when its related navigation was aborted.
This adds a handler to the `NavigationInterceptOptions` when we are
intercepting a `NavigateEvent`. This means that the scroll and focus
restoration will be delayed until the handler promise resolves. It also
means that we can provide better indication of an ongoing navigation
event.
This further builds out the Router integration with the platform
Navigation API. Key features in this state include:
* History restoration via direct platform APIs rather than markers left on
`history.state`. This means more guaranteed correctness and less
internal code to compute traversal restorations.
* Ability to observe navigations triggered outside the Router APIs.
Practically speaking, this means some navigations can be performed
through the platform rather than requiring `Router.navigate`. Note
that because the `NavigateEvent` is never intercepted at this point of
the implementation, regular anchor tags cannot be used because they
will still trigger a popstate navigation.
This implementation does _not_ intercept the `NavigateEvent` but future
iterations should. By omitting the interception, we are missing out on
features such as:
* Platform-supported scroll and focus reset
* Holding the navigate event open for the duration of the router
navigation, allowing for a visual loading indicator in the browser
* Support for intercepting navigations from regular anchor tags (e.g.
not `RouterLink`s) and converting those to SPA navigations.
PR Close#64905
Matthieu Riegler