With the directive-based control flow users were able to conditionally project content using the `*` syntax. E.g. `<div *ngIf="expr" projectMe></div>` will be projected into `<ng-content select="[projectMe]"/>`, because the attributes and tag name from the `div` are copied to the template via the template creation instruction. With `@if` and `@for` that is not the case, because the conditional is placed *around* elements, rather than *on* them. The result is that content projection won't work in the same way if a user converts from `*ngIf` to `@if`.
These changes aim to cover the most common case by doing the same copying when a control flow node has *one and only one* root element or template node.
This approach comes with some caveats:
1. As soon as any other node is added to the root, the copying behavior won't work anymore. A diagnostic will be added to flag cases like this and to explain how to work around it.
2. If `preserveWhitespaces` is enabled, it's very likely that indentation will break this workaround, because it'll include an additional text node as the first child. We can work around it here, but in a discussion it was decided not to, because the user explicitly opted into preserving the whitespace and we would have to drop it from the generated code. The diagnostic mentioned point #1 will flag such cases to users.
Fixes#52277.
PR Close#52414
The previous commits provided the scaffolding for `defer on`. In this commit, we build on that work, adding triggers for `immediate`, `timer`, `hover`, and `viewport`.
PR Close#52387
Previously, we supported a `HasConst` trait, allowing an op to be const collected automatically. However, that approach had the shortcoming that each op could only collect a single constant.
Instead, we now provide a `ConstCollectedExpr`, which collects constants at the expression level, allowing ops to have multiple collectible consts.
Then, we use this new abstraction to support the `defer on` conditions.
PR Close#52387
Previously, we had an "empty shell" implementation of defer conditions, and we used separate ops to represent secondary defer blocks.
Now, we have a real scaffolding for supporting the various defer conditions, and the secondary defer block information has been refactored onto the main defer op.
Additionally, to enable this, we refactor the way that using slot indices works. Instead of having a trait that causes users of slot indices to be linked to the allocated slot, we share a single `SlotHandle` object by reference. This allows an op to use slot information for more than one Xref at a time, and eliminates a layer of indirection.
Co-authored-by: Alex Rickabaugh <alxhub@users.noreply.github.com>
PR Close#52387
Enables a handful of i18n tests that are currently skipped, but pass if
enabled. Some of them require alternate golden files because of
inconsequential differences in the cost array order.
PR Close#52390
Previously, we would emit *two* pipe creation instructions for each pipe in a switch case. This is because we were visiting both the transformed and raw versions of the pipe bindings.
Now, we clear the raw case expressions array after generating the transformed test expression.
Also, we introduce some new goldens, because our pipe creation order is harmlessly different.
PR Close#52289
We roughly attempt to match TemplateDefinitionBuilder's pipe creation order, by placing pipe creation instructions after their target elements. However, we cannot fully emulate the "inside-out" ordering TemplateDefinitionBuilder uses when multiple pipes apply to one element, because TemplateDefinitionBuilder creates the pipes as expressions are visited, from the leaves up. Our order is perfectly adequate though.
We also add a non-compatibility-mode ordering, which just appends them to the end of the create block. This is better because it allows for more chaining opportunities.
PR Close#52289
Singleton property interpolation instructions consume only one variable, but are still emitted as an interpolation instruction (they cannot be collapsed because `propertyInterpolate` implicitly stringifies its argument.)
PR Close#52289
We were incorrectly emiting a extracted constant pool index for the final argument of the projection instruction. It actually takes an array literal.
(N.B.: This means we re-create the array every time! We should probably modify the runtime to use a const index for this.)
Additionally, we alter the projection op to not extend the element op base type.
PR Close#52289
The correct order of attributes and properties is:
1. Interpolated properties
2. Interpolated attributes
3. Non-interpolated properties
4. Non-interpolated attributes
This includes an additional nuance: singleton attribute interpolations, such as `[attr.foo]="{{bar}}"`, will be "collaped" into a simple `attribute` instruction. However, this is *not* the case for singleton property interpolations! The ordering phase must take this nuance into account to match the TemplateDefinitionBuilder order.
After the project lands, it might be nice to also collapse singleton property interpolations.
PR Close#52289
Previously, we ran the ordering phase near the end of the compilation. However, this meant that phases like slot assignment and variable offset assignment would happen first, and then the nice, monotonically-increasing orders would be scrambled by the reordering.
It's much more intelligible to order first, and then perform these assignments. However, to make this happen, some modifications to the ordering phase are required. In particular, we can no longer rely on `advance` instructions to break up orderable groups.
PR Close#52289
Many instructions consume variable slots, which are used to persist data between update runs. For top-level instructions, the offset into the variable data array is implicitly advanced, because those instructions always run.
However, instructions in non-top-level expressions cannot be assumed to run every time, because they might be conditionally executed. Therefore, they cannot implicitly advance the offset into the variable data, and must be given an explicitly assigned variable offset.
TemplateDefinitionBuilder assigned offsets top-to-bottom for all instructions *except* pure functions. Pure functions would be assigned offsets lazily, on a second pass.
Template Pipeline can now imitate this behavior, when in compatibility mode: pure functions are assigned offsets on a second pass.
This also makes the "variadic var offsets" phase unnecessary -- the new approach is more general and correct.
PR Close#52289
Previously, inside an event listener, template pipeline would always save the context from restoring a view, e.g.
```
const restored_ctx = r0.ɵɵrestoreView(s);
```
This is usually correct! However, consider the case of a listener in the template's root view. The appropriate context will already be available via closure capture, and we can just use it (as `ctx`).
Now, the context resolution phase understands that we don't need to use the restored view's saved context if we would have access to it by closure.
Note: we also create a new golden, because the const array is in a harmlessly different order.
PR Close#52289
Previously, the template pipeline did not handle "empty" reads gracefully: it would emit syntactically invalid reads of empty properties. Now we read `$implicit`.
This allows us to enable a test that relies on `$implicit`. However, we also have to create another golden, because our variable inlining is more aggressive.
PR Close#52289
We currently allow elements to be collapsed around pipe creation instructions. TemplateDefinitionBuilder disallows this, but only sometimes. Collapsing in this case is actually less generated code, and it's OK to allow it.
PR Close#52289
The template pipeline can now generate track functions, and extract them into the constant pool (or optimize them if needed). Additionally, context variables such as `$index` can be used inside track functions and for loop bodies.
PR Close#52001
Add support for `repeaterCreate` and `repeater` instructions. Correctly count decls and vars, and support primary and empty blocks.
`track` functions are not yet extracted.
PR Close#52001
ICUs can be used outside of an i18n block. In this case the ICU should
be automatically wrapped in a new i18n block. This commit adds a new
phase to handle wrapping these bare ICUs.
PR Close#52250
ICU params in i18n messages are now resolved in the post-processing call
rather than in the initial message creation. This matches the output
generated by TemplateDefinitionBuilder.
PR Close#52250
ICUs are now ingested by adding ops to both the creation and update IR.
Both of these ops are ultimately removed before reification, but they
are needed to coordinate and link data between the creation and update
ops. This is done in a new ICU extraction phase that removes both ICU
ops and adds an i18nExpr op to the update IR.
PR Close#52250
Placing a structural directive on an element with an `i18n` attribute
was generating too many i18n blocks. This was due to both the element
and the template generating their own i18n block. To fix the issue, we
no longer generate top-level i18n blocks for structural directive
templates.
PR Close#52202
Structural directives on an ng-template (e.g. <ng-template *ngIf>) were
being assigned the wrong tag name ('ng-template' instead of null).
PR Close#52202
Fixes handling of placeholders for self-closing tags. Self-closing tags
set a combined value for the start tag placeholder, rather than separate
values for the start and close placeholders.
This commit also enables a number of now passing tests. For some of
these tests I had create a separate golden file due to the different
ordering of the const array. In the template pipeline, i18n and
attribute const collection happen in different pahses and we therefore
get a different order than TemplateDefinitionBuilder, which collected
everything in one pass. The order should not affect the overall behavior.
PR Close#52195
The way we were propagating params up to parent i18n ops didn't account
for the fact that a parent and child could both have a value for the
same placeholder. In order to properly merge the value for these cases,
we need to propagate the params up *before* serialization. Therefore I
removed the standalone param propagation phase and folded the logic into
the placeholder resolution phase.
PR Close#52195
Currently the compiler allocates a variable slot to the `@for` loop expression which ends up unused since we don't store the result on the `LView`.
PR Close#52158
Fixes that the new block syntax was generating instructions in the wrong order which meant that pipes were being declared too early. This meant that if the block is first in the template, any pipes used in it won't be able to inject things like `ChangeDetectorRef`.
These changes update the compiler and add a bunch of tests to ensure that pipes work as expected.
Fixes#52102.
PR Close#52112
This commit updates `@defer` logic related to handling `after` and `minimum` parameters tree-shakable.
If `after` or `minimum` was used on a `@loading` or `@placeholder` blocks, compiler generates an extra argument for the `ɵɵdefer` instruction. This extra argument is a reference to a function that brings timer-related code.
PR Close#52042
A new statement will be generated for components which will attach some useful debug info to them to be used in runtime error handling. Currently this only happens in full and local compilation modes.
PR Close#51919
The custom logic in the generate advance phase for i18n expressions did
not work in all cases. Instead we add a new phase to update the
expression's target op, and then allow the standard advance generation
code to determine the number of advance instructions needed.
Co-authored-by: Dylan Hunn <dylhunn@users.noreply.github.com>
PR Close#51988
Fixes that the compiler wasn't picking up pipes used inside defer block triggers as dependencies. We had implemented the `visitDeferredTrigger` visitor method, but it wasn't being called, because we weren't going through the `visitAll` method of the deferred block. We don't use `visitAll`, because child nodes have to be processed differently than the connected blocks and triggers.
Fixes#52068.
PR Close#52071
The `_enabledBlockTypes` config option was removed recently, since we've enabled @-syntax by default. This commit removes `_enabledBlockTypes` references from the `compiler-cli` test cases.
PR Close#52066
A couple tests were already passing, and just needed to be enabled. This includes tests pertaining to:
* ng-template
* host binding styling slots
* and host animation bindings
* some literal tests (which were missing some $foo$ escaped names)
We add pipeline-specific versions of the following tests, and enable them:
* A local refs test. The consts for the element attributes and the consts for local reference are collected in the reverse order, but the emitted template is functionally the same.
* A safe accesstest. Consider the expression `$any(val)?.foo`. `TemplateDefinitionBuilder` extracts a temporary variable: `($tmp_0_0$ = $ctx$.val) == null ? null : $tmp_0_0$.foo`. It presumably does this because it considers the `$any(...)` to be a function call. However, this is not a real call, so Template Pipeline safely ignores it and declines to generate a temporary.
* Another local refs test. AttributeMarker.Template is emitted at the end of the const array (instead of the middle)
PR Close#51950
Consider an `ng-template` which is generated as a result of a structural directive:
```
<div *ngFor="let inner of items"
(click)="onClick(inner)"
[title]="getTitle()"
>
```
This should logically expand into something like the following:
```
<ng-template [ngForOf]="..." >
<div (click)="..." [title]="..."></div>
</ng-template>
```
Note that the `(click)` handler and the `[title]` property are only present on the inner div, *not* on the enclosing generated `ng-template`.
Previously, Template Pipeline would place these bindings on *both* the tempate and the inner element.
However, we can't just remove them completely, because these bindings should still be matchable on the generated `ng-template` (which is very surprising, but nonetheless true).
We resolve this issue with two improvements:
(1) The ingestion step is now much smarter about determining not only if a binding is on a template element, but whether it actually targets that template element.
(2) We use `ExtractedAttributeOp` directly, rather than going through `BindingOp`, to cause the `ng-template` to still receive these bindings in its `consts` array for matching purposes.
PR Close#51950
For components, the parser already extracts the `important` property (and it is later disregarded). However, because host bindings use a totally separate parsing code path, this was never happing for host bindings.
Here, we add some code to the host style parsing phase to drop the `!important` suffix.
We could solve this category of problems for good by parsing host bindings with the same code as template bindings.
PR Close#51950
Previously, we always generated temporary variable declarations at the beginning of each view's update block. This is wrong, for two reasons:
1. Temporaries can be used in the create block
2. When listeners use temporaries, we should declare them inside the listener.
Now, we always place temporaries at the beginning of the enclosing OpList, and recursively try to generate them when we find a listener.
PR Close#51950
Reworks a few more places to output arrow functions instead of function declarations in order to reduce the amount of code we generate. Some of these places include:
* Factories in injectable definitions.
* Forward references.
* `dependencies` function in the component definition.
* `consts` function in the component definition.
PR Close#52010
The template pipeline now supports basic forms of `defer` blocks. This includes the `loading`, `placeholder`, and `error` blocks, as well as the loading and placeholder configuration options.
Lazy dependencies and prefetch are not yet implemented.
PR Close#51942
Enables the new `@` block syntax by default by removing the `enabledBlockTypes` flags. There are still some internal flags that allow special use cases to opt out of the block syntax, like during XML parsing and when compiling older libraries (see #51979).
PR Close#51994
Increases the `minVersion` of component declarations that use bloks to v17 in order to indicate to users that they need to update if the library they're using is on the new syntax, while preserving backwards compatibility for libraries that do not use the syntax.
PR Close#51979
We were previously emitting pure functions as `function foo(args) {return bar;}`, but `TemplateDefinitionBuilder` uses arrow functions instead (`const foo = (args) => bar`). By matching this behavior, we can enable many additional tests.
PR Close#51961
This is a deceptively simple fix for a deep issue. Consider the following template:
```
<button [title]="myTitle" [id]="(auth().identity() | async)" [tabindex]="1">
```
`TemplateDefinitionBuilder` allocates the following variable (binding) slots:
v[0] = [title] binding
v[1] = [id] binding
v[2] = [tabindex] binding
v[3] = pipe binding
v[4] = pipe binding
As you can see, all three top-level property bindings were assigned variable indices. Then, variables for nested expressions were assigned.
Before this change, Template Pipeline would choose the following order:
v[0] = [title] binding
v[1] = [id] binding
v[2] = pipe binding
v[3] = pipe binding
v[4] = [tabindex] binding
With this order, nested expressions have their variables counted and assigned before subsequent top-level property bindings. This results in different variable indices for `pipeBinding` expressions that are not inside the final property binding.
However, this is not just different -- it's actually incorrect! Consider a case like the following:
```
<button [p1]="c ? (a | pipe) : 3" [p2]="b | pipe">
```
These pipe bindings are executed *conditionally*. This means that, because we don't count and assign all the "fixed" variable slots first, i.e. those belonging to the property bindings, their indices might end up incorrect, depending on whether or not a pipeBinding happened as part of the update block.
With this change, we count all variables on top-level ops first, and then descend into all expressions.
PR Close#51961
An `if` block can specify an alias for its main expression. We now support these in the template pipeline:
- We generate a temporary variable for the original expression
- We pass the temporary to the `conditional` instruction's context argument
- We provide the alias's name in the ambient context variables map
The context variables map now also accepts a name whose lookup value on the context object is empty. This will be interpreted as a read of the entire context object.
PR Close#51931
This entails adding a bit of extra logic to the existing conditional ingestion and corresponding phase, because `if` blocks lack a test expression.
Additionally, enable a couple more `switch` tests by resolving a curious issue -- we now consume a variable for conditionals.
PR Close#51931
