Introduce ValidateMessage to enable alternative protobuf runtimes by andrewparmet · Pull Request #480 · bufbuild/protovalidate-java

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

andrewparmet · 2026-06-14T15:17:46Z

Adds a runtime-independent ValidateMessage interface so alternative protobuf runtimes (for example, protokt) can be validated without first converting their messages to com.google.protobuf.Message.

Changes

Add the public ValidateMessage and Value surface: getDescriptorForType, hasField, getField, celValue, plus rawValue / jvmValue for the value side.
Add ProtobufValidateMessage, which adapts com.google.protobuf.Message to ValidateMessage, so the existing Message validation path is unchanged.
Route the evaluators (FieldEvaluator, OneofEvaluator, MessageOneofEvaluator, AnyEvaluator, MapEvaluator, list/object values, WrappedValueEvaluator) through ValidateMessage instead of Message directly.
Expose Validator.validate(ValidateMessage) as the entry point for alternative-runtime callers; the Message overload delegates through it.

Related work

protovalidate-java: Use the CEL planner runtime #481 moves the CEL setup to the planner runtime; the planner is what protokt's ValidateMessage implementation runs against.
protokt: Implement lazy message conversion for protovalidate open-toast/protokt#402 is the downstream consumer that validates protokt messages through ValidateMessage.
Supersedes Introduce MessageReflector to enable alternative protobuf runtimes #202, an earlier take on the same ValidateMessage interface.

Testing

./gradlew :conformance:conformance passes.

Moves protovalidate-java's CEL setup from the deprecated standard runtime to the planner runtime. ## Notable callouts - Wire the compiler and runtime by hand rather than through `plannerCelBuilder()`. `CelRuntimeImpl` directs callers to subset the standard library via `setStandardFunctions` rather than `setStandardEnvironmentEnabled`, so the runtime drops stdlib `matches` that way while the checker continues to use `setStandardEnvironmentEnabled(false)`. This keeps `CustomOverload`'s caching `matches` replacement (cel-java#1038). - Set `CelOptions.current().enableHeterogeneousNumericComparisons(true)`, which `CelRuntimeImpl` requires. - Update `cel-java` to 0.13.1 preparing to support alternative Protobuf runtimes. ## Related work: - cel-java: 0.13.1 carries the CelValue-path fixes this work relies on (fixed32/64 treated as unsigned; FieldMask navigable as a message, cel-expr/cel-java#1074). - protovalidate-java: #480 introduces `MessageReflector` to support alternative Protobuf runtimes, which builds on this runtime change (supersedes #202). - protokt: open-toast/protokt#402 is the downstream consumer that validates protokt messages through `MessageReflector` on this planner runtime. ## Testing `./gradlew :conformance:conformance` passes.

jonbodner-buf · 2026-06-17T16:25:50Z

@andrewparmet can you update the branch from main?

andrewparmet · 2026-06-17T16:28:45Z

@jonbodner-buf done!

jonbodner-buf · 2026-06-17T17:34:19Z

My biggest concern is there are no tests to validate these changes. Yes, the existing tests still pass, but nothing in the existing tests calls public ValidationResult validate(MessageReflector message) throws ValidationException .

The exception thrown by build.buf.protovalidate.MessageValue#jvmValue and build.buf.protovalidate.MessageReflectorValue#jvmValue also concern me, want to make sure that there is no code path where that could be triggered, especially for MessageReflectorValue, since it will be used with external types that implement MessageReflector.

Rename the MessageReflector interface to ValidateMessage, and accordingly ProtobufMessageReflector -> ProtobufValidateMessage and MessageReflectorValue -> ValidateMessageValue (with its `reflector` field/locals renamed to `message`). Also give the jvmValue UnsupportedOperationException on the message-backed Value types a descriptive message: MessageValue notes it carries the rule message and should be read via messageValue()/celValue(); the ValidateMessage-backed value notes that a getField() implementation returned a message value for a field evaluated as a scalar.

Add ValidateMessageTest, which validates through a map-backed ValidateMessage (no com.google.protobuf.Message): scalar / nested / repeated / map / repeated-message / map-message violations and their field paths, a valid message, the well-known Timestamp type, bytes, enum, unsigned (uint32), the Int32Value wrapper, a required oneof, the clear error when a message field is not backed by a ValidateMessage, and that jvmValue is unsupported on the message-backed Value types. Co-authored-by: Jon Bodner <jbodner@buf.build>

andrewparmet · 2026-06-18T03:21:14Z

Pushed up some tests - thanks for the starter patch from your investigation! The existing Validator tests do indirectly call validate(ValidateMessage message) via the Message overload, which hands off directly to it.

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Introduce MessageReflector to enable alternative protobuf runtimes
5624a94

andrewparmet mentioned this pull request Jun 14, 2026
Use the CEL planner runtime #481
Merged

andrewparmet marked this pull request as ready for review June 14, 2026 15:22

andrewparmet added 2 commits June 15, 2026 15:23

Merge branch 'main' into pv-message-reflector-pure
7ae533c

Apply spotless formatting
3788308

Merge branch 'main' into pv-message-reflector-pure
8ceebfd

andrewparmetand others added 2 commits June 17, 2026 23:15

andrewparmet changed the title ~~Introduce MessageReflector to enable alternative protobuf runtimes~~Introduce ValidateMessage to enable alternative protobuf runtimesJun 18, 2026

Uh oh!

Conversation

andrewparmet commented Jun 14, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes

Related work

Testing

Uh oh!

jonbodner-buf commented Jun 17, 2026

Uh oh!

andrewparmet commented Jun 17, 2026

Uh oh!

jonbodner-buf commented Jun 17, 2026

Uh oh!

andrewparmet commented Jun 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

andrewparmet commented Jun 14, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes

Related work

Testing

Uh oh!

jonbodner-buf commented Jun 17, 2026

Uh oh!

andrewparmet commented Jun 17, 2026

Uh oh!

jonbodner-buf commented Jun 17, 2026

Uh oh!

andrewparmet commented Jun 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

andrewparmet commented Jun 14, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes

Related work

Testing

Uh oh!

jonbodner-buf commented Jun 17, 2026

Uh oh!

andrewparmet commented Jun 17, 2026

Uh oh!

jonbodner-buf commented Jun 17, 2026

Uh oh!

andrewparmet commented Jun 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

andrewparmet commented Jun 14, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changes

Related work

Testing

Uh oh!

andrewparmet commented Jun 14, 2026•
edited
Loading

andrewparmet commented Jun 14, 2026•
edited
Loading

andrewparmet commented Jun 14, 2026•
edited
Loading

andrewparmet commented Jun 14, 2026•
edited
Loading

andrewparmet commented Jun 14, 2026•
edited
Loading