With #28594 we refactored the `@angular/compiler` slightly to
allow opting out from external symbol re-exports which are
enabled by default.
Since symbol re-exports only benefit projects which have a
very strict dependency enforcement, external symbols should
not be re-exported by default as this could grow the size of
factory files and cause unexpected behavior with Angular's
AOT symbol resolving (e.g. see: #25644).
Note that the common strict dependency enforcement for source
files does still work with external symbol re-exports disabled,
but there are also strict dependency checks that enforce strict
module dependencies also for _generated files_ (such as the
ngfactory files). This is how Google3 manages it's dependencies
and therefore external symbol re-exports need to be enabled within
Google3.
Also "ngtsc" also does not provide any way of using external symbol
re-exports, so this means that with this change, NGC can partially
match the behavior of "ngtsc" then (unless explicitly opted-out).
As mentioned before, internally at Google symbol re-exports need to
be still enabled, so the `ng_module` Bazel rule will enable the symbol
re-exports by default when running within Blaze.
Fixes#25644.
PR Close#28633
Users might have run the CSS Preprocessor tool *before* the Angular
compiler. For example, we do it that way under Bazel. This means that
the design-time reference is different from the compile-time one - the
input to the Angular compiler is a plain .css file.
We assume that the preprocessor does a trivial 1:1 mapping using the same
basename with a different extension.
PR Close#28166
This commit adds tracking of modules, directives, and pipes which are made
visible to consumers through NgModules exported from the package entrypoint.
ngtsc will now produce a diagnostic if such classes are not themselves
exported via the entrypoint (as this is a requirement for downstream
consumers to use them with Ivy).
To accomplish this, a graph of references is created and populated via the
ReferencesRegistry. Symbols exported via the package entrypoint are compared
against the graph to determine if any publicly visible symbols are not
properly exported. Diagnostics are produced for each one which also show the
path by which they become visible.
This commit also introduces a diagnostic (instead of a hard compiler crash)
if an entrypoint file cannot be correctly determined.
PR Close#27743
In a more specific scenario: Considering people use a custom TypeScript compiler host with `NGC`, they _could_ expect only posix paths in methods like `writeFile`. This at first glance sounds like a trivial issue that should be just fixed by the actual compiler host, but usually TypeScript internal API's just pass around posix normalized paths, and therefore it would be good to follow the same standards when passing JSON genfiles to the `CompilerHost`.
For normal TypeScript files (and TS genfiles), this is already consistent because those will be handled by the actual TypeScript `Program` (see `emitCallback`).
PR Close#27062
Originally, the ivy_switch mechanism used Bazel genrules to conditionally
compile one TS file or another depending on whether ngc or ngtsc was the
selected compiler. This was done because we wanted to avoid importing
certain modules (and thus pulling them into the build) if Ivy was on or
off. This mechanism had a major drawback: ivy_switch became a bottleneck
in the import graph, as it both imports from many places in the codebase
and is imported by many modules in the codebase. This frequently resulted
in cyclic imports which caused issues both with TS and Closure compilation.
It turns out ngcc needs both code paths in the bundle to perform the switch
during its operation anyway, so import switching was later abandoned. This
means that there's no real reason why the ivy_switch mechanism needed to
operate at the Bazel level, and for the ivy_switch file to be a bottleneck.
This commit removes the Bazel-level ivy_switch mechanism, and introduces
an additional TypeScript transform in ngtsc (and the pass-through tsc
compiler used for testing JIT) to perform the same operation that ngcc
does, and flip the switch during ngtsc compilation. This allows the
ivy_switch file to be removed, and the individual switches to be located
directly next to their consumers in the codebase, greatly mitigating the
circular import issues and making the mechanism much easier to use.
As part of this commit, the tag for marking switched variables was changed
from __PRE_NGCC__ to __PRE_R3__, since it's no longer just ngcc which
flips these tags. Most variables were renamed from R3_* to SWITCH_* as well,
since they're referenced mostly in render2 code.
Test strategy: existing test coverage is more than sufficient - if this
didn't work correctly it would break the hello world and todo apps.
PR Close#26550
Closure requires @nocollapse on Ivy definition static fields in order
to not convert them to standalone constants. However tsickle, the tool
which would ordinarily be responsible for adding @nocollapse, doesn't
properly annotate fields which are added synthetically via transforms.
So this commit adds @nocollapse by applying regular expressions against
code during the final write to disk.
PR Close#25775
This commit takes the first steps towards ngtsc producing real
TypeScript diagnostics instead of simply throwing errors when
encountering incorrect code.
A new class is introduced, FatalDiagnosticError, which can be thrown by
handlers whenever a condition in the code is encountered which by
necessity prevents the class from being compiled. This error type is
convertable to a ts.Diagnostic which represents the type and source of
the error.
Error codes are introduced for Angular errors, and are prefixed with -99
(so error code 1001 becomes -991001) to distinguish them from other TS
errors.
A function is provided which will read TS diagnostic output and convert
the TS errors to NG errors if they match this negative error code
format.
PR Close#25647
In tsc 3.0 the check that enables program structure reuse in tryReuseStructureFromOldProgram has changed
and now uses identity comparison on arrays within CompilerOptions. Since we recreate the options
on each incremental compilation, we now fail this check.
After this change the default set of options is reused in between incremental compilations, but we still
allow options to be overriden if needed.
PR Close#25275
The optional property on `ts.CompilerHost` is called `realpath` (lower
case), not `realPath` (lower camel case).
It is not clear to me what the impact of this is, but the author's
intent was clearly to override `realpath`.
PR Close#25023
ngtsc used to have a custom ts.CompilerHost which delegated to the plain
ts.CompilerHost. There's no need for this wrapper class and it causes
issues with CLI integration, so delete it.
PR Close#25080
Previously ngtsc would use a tuple of class types for listing metadata
in .d.ts files. For example, an @NgModule's declarations might be
represented with the type:
[NgIf, NgForOf, NgClass]
If the module had no declarations, an empty tuple [] would be produced.
This has two problems.
1. If the class type has generic type parameters, TypeScript will
complain that they're not provided.
2. The empty tuple type is not actually legal.
This commit addresses both problems.
1. Class types are now represented using the `typeof` operator, so the
above declarations would be represented as:
[typeof NgIf, typeof NgForOf, typeof NgClass].
Since typeof operates on a value, it doesn't require generic type
arguments.
2. Instead of an empty tuple, `never` is used to indicate no metadata.
PR Close#24862
All errors for existing fields have been detected and suppressed with a
`!` assertion.
Issue/24571 is tracking proper clean up of those instances.
One-line change required in ivy/compilation.ts, because it appears that
the new syntax causes tsickle emitted node to no longer track their
original sourceFiles.
PR Close#24572
Bazel has a restriction that a single output (eg. a compiled version of
//packages/common) can only be produced by a single rule. This precludes
the Angular repo from having multiple rules that build the same code. And
the complexity of having a single rule produce multiple outputs (eg. an
ngc-compiled version of //packages/common and an Ivy-enabled version) is
too high.
Additionally, the Angular repo has lots of existing tests which could be
executed as-is under Ivy. Such testing is very valuable, and it would be
nice to share not only the code, but the dependency graph / build config
as well.
Thus, this change introduces a --define flag 'compile' with three potential
values. When --define=compile=X is set, the entire build system runs in a
particular mode - the behavior of all existing targets is controlled by
the flag. This allows us to reuse our entire build structure for testing
in a variety of different manners. The flag has three possible settings:
* legacy (the default): the traditional View Engine (ngc) build
* local: runs the prototype ngtsc compiler, which does not rely on global
analysis
* jit: runs ngtsc in a mode which executes tsickle, but excludes the
Angular related transforms, which approximates the behavior of plain
tsc. This allows the main packages such as common to be tested with
the JIT compiler.
Additionally, the ivy_ng_module() rule still exists and runs ngc in a mode
where Ivy-compiled output is produced from global analysis information, as
a stopgap while ngtsc is being developed.
PR Close#24056
This commit adds a new compiler pipeline that isn't dependent on global
analysis, referred to as 'ngtsc'. This new compiler is accessed by
running ngc with "enableIvy" set to "ngtsc". It reuses the same initialization
logic but creates a new implementation of Program which does not perform the
global-level analysis that AngularCompilerProgram does. It will be the
foundation for the production Ivy compiler.
PR Close#23455
rxjs 6.0.0 breaks strictMetadataEmit as they now publish a .d.ts file with a
structure like:
declare export class Subscription {
static EMPTY: Subscription;
}
This generates metadata which contains an error, and fails the strictMetadataEmit
validation. There is nothing a library author can do in this situation except to
set strictMetadataEmit to false.
The spirit of strictMetadataEmit is to validate that the author's library doesn't
do anything that will break downstream users. This failure is a corner case which
causes more harm than good, so this commit disables validation for metadata
collected from .d.ts files.
Fixes#22210
PR Close#23275
Lowering expressions in flat module metadata is desirable, but it won't
work without some rearchitecting. Currently the flat module index source
is added to the Program and therefore must be determined before the rest
of the transforms run. Since the lowering transform changes the set of
exports needed in the index, this creates a catch-22 in the index
generation.
This commit causes the flat module index metadata to be generated using
only those transforms which are "safe" (don't modify the index).
PR Close#23226
Currently, the flat module index metadata is produced directly from
the source metadata. The compiler, however, applies transformations
on the Typescript sources during transpilation, and also equivalent
transformations on the metadata itself. This transformed metadata
doesn't end up in the flat module index.
This changes the compiler to generate the flat module index metadata
from its transformed version instead of directly from source.
PR Close#23129
Computing the value of loadChildren does not work externally, as the CLI
needs to be able to detect the paths referenced to properly set up
codesplitting. However, internally, different approaches to codesplitting
require hashed module IDs, and the computation of those hashes involves
something like:
{path: '...', loadChildren: hashFn('module')}
ngc should lower loadChildren into an exported constant in that case.
This will never break externally, because loadChildren is always a
string externally, and a string won't get lowered.
PR Close#23088
This allows a bundle index to be re-exported by a higher-level module without fear of collisions.
Under bazel, we always set the prefix to be underscore-joined workspace, package, label
PR Close#23007
ngc knows to filter out d.ts inputs, but the logic accidentally
depended on whether it had a previous Program lying around.
Fixing that logic puts ngc on the fast code path, but in that code
path it must be able to merge tsickle EmitResults, so we need to
plumb the tsickle.mergeEmitResults function through all the intervening
APIs. The bulk of this change is that plumbing.
PR Close#22899
This adds compilation of @NgModule providers and imports into
ngInjectorDef statements in generated code. All @NgModule annotations
will be compiled and the @NgModule decorators removed from the
resultant js output.
All @Injectables will also be compiled in Ivy mode, and the decorator
removed.
PR Close#22458
BREAKING CHANGE:
The `<template>` tag was deprecated in Angular v4 to avoid collisions (i.e. when
using Web Components).
This commit removes support for `<template>`. `<ng-template>` should be used
instead.
BEFORE:
<!-- html template -->
<template>some template content</template>
# tsconfig.json
{
# ...
"angularCompilerOptions": {
# ...
# This option is no more supported and will have no effect
"enableLegacyTemplate": [true|false]
}
}
AFTER:
<!-- html template -->
<ng-template>some template content</ng-template>
PR Close#22783
Works around an issue with TypeScript 2.6 and 2.7 that causes
the tranformer emit to emit incorrect escapes for css string
literals.
Fixes: #22774
PR Close#22776
Closure has a transformation which turns:
Service.ngInjectableDef = ...;
into:
Service$ngInjectableDef = ...;
This transformation obviously breaks Ivy in a major way. The solution is
to annotate the fields as @nocollapse. However, Typescript appears to ignore
synthetic comments added to a node during a transformation, so the "right"
way to add these comments doesn't work.
As an interim measure, a post-processing step just before the compiled JS is
written to disk appends the correct comments with a regular expression.
PR Close#22691
Previously the flag would only disable the check in the case we tried to use newer tsc version.
In g3 we sometimes take a while to update tsc, but as a prerequisite of that Angular needs to be
updated first. This change enables us to update Angular and use it in g3 while g3 is being update
to the required tsc. Of course extra care is required when this check is disabled, but since we
control everything in g3, it's on us to get this right.
I don't see any preexisting tests for this, and I'm not sure how to write them right now.
I filed https://github.com/angular/angular/issues/22699
PR Close#22669
