chore: migrate google-cloud-aiplatform to librarian

[quirogas]

Migration of google-cloud-aiplatform to Librarian

This PR migrates the google-cloud-aiplatform package to use the Librarian generation workflow.

Warning

This PR is currently a draft and contains breaking changes (detailed below) due to the removal of veneer helper files and schema proto definitions. These issues need to be resolved before merging.

User-facing Changes

If merged as-is, this PR introduces the following breaking changes:

• Removal of Schema Proto Definitions: Proto definitions under google/cloud/aiplatform/v1/schema/... and v1beta1/schema/... are no longer compiled into the protos bundle (protos.js/protos.json). Users will no longer be able to access these types via the exported protos namespace.
• Removal of Veneer Helpers: The helpers export and the automatically injected .toValue() (prototype) and .fromValue() (static) helper methods on AI Platform schema objects are removed.

User Impact & Real-world Code References

Removing the schema proto definitions and veneer helpers will break existing user code that relies on type safety, instantiation, or value conversion for AI Platform schemas.

Below are the 3 main impacted cases, supported by real-world code references:

Case 1: Instantiating Schema Messages & using toValue()

Users instantiate schema classes directly to benefit from validation or default values, and then convert them to google.protobuf.Value using .toValue().

Impacted Code Example:

const { protos } = require('@google-cloud/aiplatform');

// ❌ RUNTIME ERROR: Cannot read properties of undefined (reading 'ImageObjectDetectionPredictionParams')
const parametersObj = new protos.google.cloud.aiplatform.v1.schema.predict.params.ImageObjectDetectionPredictionParams({});
// ❌ RUNTIME ERROR: .toValue() is not a function
const parameters = parametersObj.toValue();

Real-world References:

• Official Google Cloud Samples: predict-image-object-detection.js (Line 50)
• Community Project: detection-controller.ts (Line 42)

Case 2: TypeScript Type Annotations

TypeScript users import the protos namespace to type-annotate their variables when constructing payloads.

Impacted Code Example:

import { protos } from '@google-cloud/aiplatform';

// ❌ COMPILE ERROR: Namespace 'schema' does not exist under 'protos.google.cloud.aiplatform.v1'
const predictionParams: protos.google.cloud.aiplatform.v1.schema.predict.params.IImageClassificationPredictionParams = {
  confidenceThreshold: 0.5,
  maxPredictions: 5,
};

Case 3: Parsing Prediction Results & using fromValue()

When receiving prediction results, users use the static .fromValue() method to convert the generic google.protobuf.Value back into a typed schema object.

Impacted Code Example:

const { protos } = require('@google-cloud/aiplatform');

// ❌ RUNTIME ERROR: Cannot read properties of undefined (reading 'ImageObjectDetectionPredictionResult')
const predictionResultObj =
  protos.google.cloud.aiplatform.v1.schema.predict.prediction.ImageObjectDetectionPredictionResult.fromValue(
    predictionResultVal
  );

Real-world References:

• Official Google Cloud Samples: predict-image-object-detection.js (Line 67)

Handwritten files being removed

The following handwritten files are currently removed in this PR.

packages/google-cloud-aiplatform/src/decorator.ts

• Purpose: Dynamically enhances generated proto message classes (listed in enhanced-types.json) with .toValue() (instance method) and .fromValue() (static method) helpers.

• Last Commit: danieljbruce on 2026-04-28 (Copyright update).

• Last Logic Change: Eric Schmidt on 2021-02-12.

• Impacted Code:

  Click to view impacted user code

  import { protos } from '@google-cloud/aiplatform';
  
  const valueProto = {
    structValue: {
      fields: {
        content: { stringValue: 'example image content' }
      }
    }
  };
  
  // BROKEN: .fromValue() will no longer exist on the schema class
  const instance = protos.google.cloud.aiplatform.v1.schema.predict.instance.ImageClassificationPredictionInstance.fromValue(valueProto);
  
  const instance2 = new protos.google.cloud.aiplatform.v1.schema.predict.instance.ImageClassificationPredictionInstance({
    content: 'example image content',
  });
  
  // BROKEN: .toValue() will no longer exist on the instance
  const convertedValue = instance2.toValue();

packages/google-cloud-aiplatform/src/helpers.ts

• Purpose: Implements the toValue and fromValue wrapper functions that convert between JavaScript objects/Messages and google.protobuf.Value objects.

• Last Commit: danieljbruce on 2026-04-28 (Copyright update).

• Last Logic Change: Eric Schmidt on 2023-06-06.

• Impacted Code:

  Click to view impacted user code

  import { helpers } from '@google-cloud/aiplatform';
  
  const myObject = { foo: 'bar' };
  
  // BROKEN: 'helpers' export will be missing, and toValue/fromValue won't be accessible
  const valueProto = helpers.toValue(myObject);
  const backToObject = helpers.fromValue(valueProto);

packages/google-cloud-aiplatform/src/value-converter.ts

• Purpose: Handles the low-level conversion logic between JavaScript primitive types/objects and the protobuf Value representation (Struct, List, etc.).
• Last Commit: danieljbruce on 2026-04-28 (Copyright update).
• Last Logic Change: Eric Schmidt on 2023-06-06.

---

Why Librarian does not process the deleted protos

The proto files under google/cloud/aiplatform/v[1|v1beta1]/schema/... are being deleted because the Librarian tool does not recursively scan subdirectories of the configured proto path by default.

1. Non-recursive discovery: In Librarian's protobuf utility (input_files.go), the directory walker is restricted to maxDepth = 1. Any subdirectory (like schema/ under v1/) is skipped:

   depth := strings.Count(strings.TrimPrefix(path, source), "/")
   if info.IsDir() && depth >= maxDepth {
       return filepath.SkipDir // Skips subdirectories
   }

2. Configuration Limitation: Because they are skipped, these auxiliary schema protos (which do not contain gRPC services themselves) must be explicitly added to additional_protos in the package configuration.
3. Current Workaround: The NodejsAPI configuration struct (language.go) only supports a list of individual file paths for additional_protos. It does not support directory paths, meaning all schema proto files must be listed individually in the config to be preserved.

[gemini-code-assist]

Code Review

This pull request updates the library configuration in .repo-metadata.json from GAPIC_COMBO to GAPIC_AUTO, which has resulted in the deletion of all handwritten helper files, tests, and schemas. The reviewer points out that manually editing generated client configuration files is prohibited and that removing these handwritten helper files is a major breaking change that must be reverted upstream in the generator.

[gemini-code-assist]

Do not manually edit generated client configuration JSON files. Any changes or updates to .repo-metadata.json must be made upstream in the generator. Additionally, changing the library_type from GAPIC_COMBO to GAPIC_AUTO has resulted in the deletion of all handwritten helper files (helpers.ts, decorator.ts, value-converter.ts) and their tests, which is a major breaking change. Please revert this change upstream in the generator.

References

1. Do not manually edit generated client configuration JSON files. Any changes or updates must be made upstream in the generator.

[jskeet]

See googleapis/librarian#6280 for why this is deleting so much - I think we need to discuss it in more detail.

[suztomo]

I enriched the pull request description.

[suztomo]

Jon's #8666 is a new approach.