… deprecated.

With the minimum AGP required to be 7.0 the option is no longer relevant and its code path has been removed.

Also added a bit more documentation to the other options.

RELNOTES=Deprecate the `enableTransformForLocalTests` in the Hilt Android Gradle plugin.
PiperOrigin-RevId: 544721475

This adds supports for SPI Plugins used in KSP to be discoverable during ComponentProcessor's run in the dedicate Dagger compile task when aggregation is turned ON.

RELNOTES=N/A
PiperOrigin-RevId: 544737396

…ule.

RELNOTES=n/a
PiperOrigin-RevId: 544783996

… class name of the annotations on the synthetic property.

This prevents XProcessing from needing to wrap unnecessary annotations in `XAnnotations.equivalence()` which can throw a `TypeNotPresentException` for annotations that have error types in their annotation values.

RELNOTES=N/A
PiperOrigin-RevId: 546089941

RELNOTES=N/A
PiperOrigin-RevId: 546857277

This version of the jar contains the updated kotlinx-metadata-jvm needed to support Kotlin 1.9.

Fixes #3951

RELNOTES=N/A
PiperOrigin-RevId: 547249179

RELNOTES=n/a
PiperOrigin-RevId: 547330693

The flag is off by default but can be enabled by compiling with the following annotation processor options: `-Adagger.ignoreProvisionKeyWildcards=enabled`.

When enabled, Dagger's annotation processor will no longer allow provisioning two bindings that only differ by the types in their wild cards.

See https://dagger.dev/dev-guide/compiler-options#ignore-provision-key-wildcards

RELNOTES=Release `dagger.ignoreProvisionKeyWildcards` flag (See https://dagger.dev/dev-guide/compiler-options#ignore-provision-key-wildcards).
PiperOrigin-RevId: 547560456

RELNOTES=N/A
PiperOrigin-RevId: 548025923

RELNOTES=N/A
PiperOrigin-RevId: 548241407

RELNOTES=N/A
PiperOrigin-RevId: 548753175

This CL fixes an issue in KSP where a `KSType` from a java source may show up as nullable due to an @nullable (TYPE_USE target) annotation. In this case, the annotation actually changes the `KSType` itself (e.g. `@Nullable Foo` becomes `Foo?`), and we ended up treating the type as nullable in Dagger. This CL fixes the issue by ignoring type nullability altogether for java sources since we know that Java sources can't actually declare a nullable type. This matches Dagger's KAPT behavior for these types.

RELNOTES=N/A
PiperOrigin-RevId: 548799879

PiperOrigin-RevId: 548820787

Closes #3913
Fixes #3967

RELNOTES=N/A
PiperOrigin-RevId: 549005035

This CL fixes an issue in KSP where a KSType from a java source may show up as nullable due to an @nullable (TYPE_USE target) annotation. In this case, the annotation actually changes the KSType itself (e.g. @nullable Foo becomes Foo?), and we ended up treating the type as nullable in Dagger. This CL fixes the issue by ignoring type nullability altogether for java sources since we know that Java sources can't actually declare a nullable type. This matches Dagger's KAPT behavior for these types.

RELNOTES=N/A
PiperOrigin-RevId: 549087723

RELNOTES=N/A
PiperOrigin-RevId: 549142634

RELNOTES=N/A
PiperOrigin-RevId: 551097100

This CL fixes an issues in our GitHub Actions where we are consistently going over our cache limit. Typically, we would need to clean out the cache manually at https://github.com/google/dagger/actions/caches. This script automates that process by deleting caches each run.

  - For a PR that is already closed: deletes all corresponding caches.
  - For duplicate caches with the same Git reference and cache key: deletes the older cache.

The script should be idempotent, so I decided to run the script both at the beginning and end of each workflow. This ensure that the caches are cleaned up after a workflow ends, but also that we don't start accumulating unnecessary caches when a workflow fails before the script can be run.

RELNOTES=N/A
PiperOrigin-RevId: 551349636

…ementation details into Dagger internal.

RELNOTES=n/a
PiperOrigin-RevId: 551628363

RELNOTES=N/A
PiperOrigin-RevId: 552540977

RELNOTES=Upgraded Kotlin to 1.9.0, KSP to 1.9.0-1.0.12, Bazel to 6.3.1, and XProcessing jar.
PiperOrigin-RevId: 553246632

…se HiltCompilerTests.

RELNOTES=n/a
PiperOrigin-RevId: 553304510

RELNOTES=N/A
PiperOrigin-RevId: 553458202

…icts with KSP's Gradle plugin.

See #3965

RELNOTES=Add a better error message when Hilt Gradle Plugin class loader conflicts with KSP's Gradle plugin.
PiperOrigin-RevId: 553467242

RELNOTES=N/A
PiperOrigin-RevId: 553491705

RELNOTES=N/A
PiperOrigin-RevId: 553495719

RELNOTES=n/a
PiperOrigin-RevId: 553561912

RELNOTES=N/A
PiperOrigin-RevId: 554507627

…h nullable types.

This CL adds an @nullable annotation in the error message when formatting binding methods with nullable types like `Foo?`. This behavior matches KAPT's generated java stubs which will add an `@Nullable` annotation to the stub's method when the return type of the method is a nullable type.

RELNOTES=N/A
PiperOrigin-RevId: 554830004

…ods in KSP.

This relies on recent support in XProcessing added in aosp/2674117. Before this change we were only correctly finding the property methods in KAPT but missing them in KSP.

RELNOTES=N/A
PiperOrigin-RevId: 555172373

RELNOTES=N/A
PiperOrigin-RevId: 555635748

RELNOTES=n/a
PiperOrigin-RevId: 556833204

RELNOTES=N/A
PiperOrigin-RevId: 557094089

This CL also flips the default for `dagger.ignoreProvisionKeyWildcards` to `ENABLED`.

RELNOTES=Require `dagger.ignoreProvisionKeyWildcards=ENABLED` when using KSP and flip the default to `ENABLED`.
PiperOrigin-RevId: 557170874

RELNOTES=n/a
PiperOrigin-RevId: 557234097

Using keywords as parameter names has been working with KAPT because KAPT just renames the parameters to something that is valid Java in the generated Java stubs. However, in order to support this in KSP we need to switch over to use `XExecutableParameterElement#jvmName` rather than `XExecutableParameterElement#name`.

This CL also requires updates from XProcessing, so I've updated Dagger's XProcessing jars as well.

Fixes #3995: Allow kotlin source to use java keywords as parameter names in KSP

RELNOTES=Fixed #3995: Allow kotlin source to use java keywords as parameter names in KSP
PiperOrigin-RevId: 557565286

RELNOTES=n/a
PiperOrigin-RevId: 557882685

RELNOTES=N/A
PiperOrigin-RevId: 558176324

RELNOTES=n/a
PiperOrigin-RevId: 559210781

RELNOTES=N/A
PiperOrigin-RevId: 559431563

RELNOTES=n/a
PiperOrigin-RevId: 559508361

RELNOTES=n/a
PiperOrigin-RevId: 559538248

This CL also updates the `javac()` and `ksp()` methods to throw if they are used with the wrong backend.

RELNOTES=N/A
PiperOrigin-RevId: 559565413

The bug in b/269172737 required us to have the scope annotation in a certain order to avoid KSP returning error types. However, it appears this ordering is no longer required with the latest Kotlin 1.9.0 and KSP 1.9.0-1.0.12.

RELNOTES=N/A
PiperOrigin-RevId: 559898050

RELNOTES=n/a
PiperOrigin-RevId: 560849062

RELNOTES=n/a
PiperOrigin-RevId: 560887022

…ted.

RELNOTES=n/a
PiperOrigin-RevId: 560887928

RELNOTES=N/A
PiperOrigin-RevId: 561732623

RELNOTES=n/a
PiperOrigin-RevId: 561788625

The issue here was that `AnnotationExpression` was using `XType#getTypeName()` resulting in `GenericType<?>.class` in generated code. This CL fixes the issue by using `XTypeElement#getClassName()` instead so that the generated code results in `GenericType.class` instead.

See #4055.

Fixes #4055

RELNOTES=Fixes #4055:
PiperOrigin-RevId: 562096148

When searching for multibinding declarations, the key (e.g. `Map<K, Provider<V>>`) has to be unwrapped since the declaration keys have the form `Map<K, V>`. This CL consolidates the logic into a single function so that we only have to document these details once.

RELNOTES=N/A
PiperOrigin-RevId: 564374943

…ull/4023/files#r1308159525 for details.

Fixes #3980.

RELNOTES=Fixes #3980.
PiperOrigin-RevId: 564502143

…ry`"

PiperOrigin-RevId: 564526719

RELNOTES=N/A
PiperOrigin-RevId: 568310453

RELNOTES=N/A
PiperOrigin-RevId: 568345537

  * Moves `abstract` methods in `ComponentDescriptor` to the top of the class so that it's easier to tell which methods contributes to the AutoValue equals/hashcode.
  * Moves `ComponentDescriptorFactory` directly to a nested class, `ComponentDescriptor.Factory`, to be more consistent with how structure other descriptors like `ModuleDescriptor`/`ModuleDescriptor.Factory`.
  * Moves `CancellationPolicy` class definition out of `ComponentDescriptor` since it's not directly related to that class.

RELNOTES=N/A
PiperOrigin-RevId: 568890232

This change should be temporary until we can get a more robust fix in KSP (google/ksp#1555).

Fixes #4063
Fixes #4054

RELNOTES=N/A
PiperOrigin-RevId: 569608460

Recently, some of our tests have started timing out on GitHub Actions (https://github.com/google/dagger/actions/runs/6355090490/job/17262572158?pr=4082):

```
//javatests/dagger/internal/codegen:ComponentProcessorTest              TIMEOUT in 315.2s
```

This CL adds the ability to shard some of our large and medium sized tests.

RELNOTES=N/A
PiperOrigin-RevId: 569752153

…nentPath.

Including the ComponentPath in ResolvedBindings avoids having to create new objects
of type ResolvedBindingsWithPath to satisfy our Map key. In addition, passing the ComponentPath while creating LegacyBindingGraph avoids having to recreate it from scratch during BindingGraphConverter phase.

RELNOTES=N/A
PiperOrigin-RevId: 570137117

RELNOTES=N/A
PiperOrigin-RevId: 570790118

This CL inlines some functions that are only used at a single call-site and are essentially 1-liners.

RELNOTES=N/A
PiperOrigin-RevId: 570810185

The logic for whether we generate a MembersInjector was duplicated in `InjectBindingRegistryImpl` and `MembersInjectorGenerator`. In particular, there are cases where `InjectBindingRegistryImpl` decided that a MembersInjector should be generated, but additional logic in `MembersInjectorGenerator` overrides that decision and decides not to generate the MembersInjector (e.g. if the class has no local injection sites).

This led to duplicating of logic in `InjectBindingRegistryImpl` when we're trying to decide if we should output a warning for the case where a MembersInjector should be generated in an upstream library because we needed to repeat the logic in `MembersInjectorGenerator` to know if the MembersInjector will actually be generated.

This CL moves the decision on whether to generate a MembersInjector completely into `InjectBindingRegistryImpl`, which removes the duplicated logic and also makes it easier to understand when the MembersInjector will actually be generated since we only need to look in one location.

This CL also removes `warnIfNotAlreadyGenerated` since `isCalledFromInjectProcessor`  contains all the information we need to decide whether to warn in this case.

RELNOTES=N/A
PiperOrigin-RevId: 571421631

RELNOTES=Add support for using `@AssistedInject` with `@HiltViewModel`.
PiperOrigin-RevId: 572216305

RELNOTES=N/A
PiperOrigin-RevId: 572362454

…() with the version in XProcessing

This also fixes a Github issue where we use `XConverters.toKS()` to get the underlying KSAnnotated and its origin but it doesn't work for synthetic elements like SyntheticExecutableParameterElement. The XProcessing version fixes this by

1. Making these functions member functions of XMemberContainer so that we can skip the implementation in most element types and we can't miss any XMemberContainer subtypes.
2. Simply consider that the member container is from Kotlin if it's KspFileMemberContainer or KspSyntheticFileMemberContainer as these are generated from Kotlin top-level properties.

RELNOTES=N/A
PiperOrigin-RevId: 572595483

RELNOTES=Add check to prevent injecting assisted factories for HiltViewModels
PiperOrigin-RevId: 572901826

…viderInstanceRequestRepresentation`.

In this case there's no need to pass in the `FrameworkInstanceSupplier` as an assisted parameter since we can derive it directly in `ProviderInstanceRequestRepresentation` using other injected objects.

RELNOTES=N/A
PiperOrigin-RevId: 573928235

This is a short-term fix for b/305748522 so that we can push through our other fix for b/302199325. The long-term fix is outlined in the b/305748522 but probably requires more design discussion.

RELNOTES=N/A
PiperOrigin-RevId: 574226829

RELNOTES=N/A
PiperOrigin-RevId: 574510675

RELNOTES=N/A
PiperOrigin-RevId: 575765188

…use `internal`.

RELNOTES=n/a
PiperOrigin-RevId: 576313767

…lag to enable it.

RELNOTES=Fixes an issue where bindings that depend on local bindings were not properly re-resolved in a component when some of its' transitive dependencies contain a cycle.
PiperOrigin-RevId: 576542469

…a flag to enable it.

RELNOTES=N/A
PiperOrigin-RevId: 576615427

There's no technical reason for Dagger's compiler to depend on Dagger Producers. In all of these cases we were pulling in producers just for either the class name or javadoc.

RELNOTES=N/A
PiperOrigin-RevId: 577017144

…CreationExtras.addCreationCallback()

These can be used to pass creation callbacks with functions like `viewModels()`.

RELNOTES=Add helper function CreationExtras.withCreationCallback() and MutableCreationExtras.addCreationCallback()
PiperOrigin-RevId: 577890589

RELNOTES=n/a
PiperOrigin-RevId: 577983612

…nd a flag to enable it.

RELNOTES=Rollforward fix for `LocalDependencyChecker#dependsOnLocalBindings` and a flag to enable it.
PiperOrigin-RevId: 578245756

This CL fixes an issues where each subcomponent was storing its complete list of accessible bindings, which includes any bindings it needs from its ancestor components. This ends up requiring much more space than actually required if we just store the bindings owned by the subcomponent and delegate to the parent component for any bindings accessible there.

I think this can likely be done for our `LegacyBindingGraph` classes as well, but I'll save that for another CL.

RELNOTES=N/A
PiperOrigin-RevId: 578306878

…y `static`/`abstract` binding methods.

Note that this doesn't remove all of the deprecated setter methods. In particular, this CL removes setter methods only for the case where a module instance is never needed because it contains only `static` and `abstract` methods.

However, this CL does not remove the deprecated setter methods in the general case where a module instance is not needed because it is unused in the component. That case is a bit trickier because it depends on the usage of the bindings rather than properties of the module itself.

RELNOTES=Potentially breaking change. Dagger no longer generates module setter methods for modules that have only static/abstract methods. These methods were previously marked as deprecated, as the module instances passed in aren't actually used by the Dagger component.
PiperOrigin-RevId: 578651221

RELNOTES=n/a
PiperOrigin-RevId: 578956204

RELNOTES=N/A
PiperOrigin-RevId: 579237165

"Bazel Test" has been the bottleneck of our builds for a while now. A large part of that is due to the default GitHub runner only having 2 cores and thus only being able to run at most 2 Bazel actions at a time. Things have also gotten worse due to the increasing number of tests in Dagger's codebase and the slowing down of tests due to the move to XProcessing Testing and possibly increased I/O due to golden files, though these issues aren't addressed in this CL.

This CL moves our "Bazel Test" GitHub actions onto the large test runner. This runner now has 16 cores which should allow 8x more tests to run in parallel and we should see a similar improvement in the time it takes these GitHub actions to run.

RELNOTES=N/A
PiperOrigin-RevId: 579881816

RELNOTES=n/a
PiperOrigin-RevId: 579923643

…nstead of `XAnnotation`s

All uses of this API immediately convert to `ClassName`, and this makes it
easier to deduplicate the returned set.

RELNOTES=N/A
PiperOrigin-RevId: 580220708

…'s ClassContext

RELNOTES=N/A
PiperOrigin-RevId: 582018651

RELNOTES=n/a
PiperOrigin-RevId: 582028428

…t contains a performance regression.

PiperOrigin-RevId: 582400758