fix: return created dataset/evaluator from create_version, not bool

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.

[papriwal]

Callers previously had to call get()/refresh() to retrieve the ARN.

Description

Fixes two issues in the AI Registry create_version API for DataSet and Evaluator:

1. DataSet.create_version() and Evaluator.create_version() returned a bare bool and silently swallowed exceptions on failure, instead of returning the newly created entity (or raising/returning None on failure). Both now return the created DataSet/Evaluator instance.

2. Adds description support to Evaluator, matching the existing DataSet pattern. description is now threaded through Evaluator.create(), refresh(), and get(), and AIRHub.import_hub_content() now forwards description to the underlying HubContentDescription API field for both entity types (previously accepted by the AWS API but never exposed by the SDK).

Also fixes a pre-existing bug in Evaluator.__init__'s super().__init__() call: it was missing a positional argument, which caused sagemaker_session to be silently written into the parent class's description slot instead of its own — meaning sagemaker_session was effectively dropped on every Evaluator instance.

Testing

• Added unit tests covering:
  • create_version returning the created entity (success) and None/raising on failure, for both DataSet and Evaluator.
  • description reaching AIRHub.import_hub_content's call args for both DataSet.create() and Evaluator.create().
  • Evaluator.get() correctly reading back description from the API response.
• Updated existing integ tests for test_create_dataset_version and test_create_evaluator_version to assert on the returned entity from create_version() instead of a boolean.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Claude finished @papriwal's task in 2m 23s —— View job

---

Code Review

I reviewed the diff (/tmp/pr.diff) against the base at origin/master. Overall this is a solid, well-tested fix. The super().__init__() correction in Evaluator is a genuine latent bug catch, and returning the created entity is a clear improvement over the old bool. A few points worth considering before merge.

1. Failure semantics diverge between the two sibling methods (main concern)

DataSet.create_version and Evaluator.create_version now handle failure in opposite ways:

• dataset.py:567-572 — catches the exception, logs it, and returns None.
• evaluator.py:535-536 — catches and re-raises as RuntimeError.

For two sibling entity types with the same method name and near-identical purpose, this is a surprising inconsistency for SDK consumers. A caller writing generic code over both types can't rely on a single contract: one signals failure with a falsy return, the other with an exception. The DataSet docstring/return type says Optional[DataSet] while Evaluator's says it always returns an Evaluator (and documents Raises: RuntimeError).

Suggest picking one convention for both. Raising is generally the more Pythonic choice (a bare None still "silently swallows" the failure detail the PR set out to fix — the caller only sees None, not why). If you keep DataSet returning None, at least call out the deliberate difference.

Note this is also a subtle backward-incompatibility: existing Evaluator callers doing if not evaluator.create_version(...): previously got False on failure and will now get an exception instead. Probably desirable, but worth being intentional about.

2. Preserve the exception chain (evaluator.py:535-536)

exceptExceptionase:
raiseRuntimeError(f"[PySDK Error] Failed to create new version: {str(e)}")

Use raise RuntimeError(...) from e so the original traceback/cause is retained. Without from e, the underlying error's context is harder to debug.

3. description isn't carried across versions (minor)

Both create_version implementations call create(...) without passing self.description, so a new version silently drops the description of the version it was derived from. Since this PR is specifically adding description support, it may be worth threading description=self.description through create_version too — otherwise the field is added but not preserved on the most common mutation path.

Nits / positives (no action needed)

• Good catch on the Evaluator.__init__super().__init__() positional-arg bug — sagemaker_session was landing in the parent's description slot. Verified against air_hub_entity.py:43-52.
• Switching from print(...) to a module logger in dataset.py is the right call.
• Test coverage for the new behavior (success/failure returns, description reaching import_hub_content, Evaluator.get() reading it back) is thorough.

I couldn't post these as inline comments — the inline-comment tool isn't available in this run — so they're consolidated here with file:line references.