JingsongLi commented on code in PR #11136:
URL: https://github.com/apache/paimon/pull/11136#discussion_r4560893621
##########
paimon-python/pypaimon/catalog/rest/rest_catalog.py:
##########
@@ -875,9 +875,10 @@ def load_table(self,
catalog_env = CatalogEnvironment(
identifier=identifier,
uuid=metadata.uuid,
- catalog_loader=self.catalog_loader(),
+ catalog_loader=None if metadata.is_external else
self.catalog_loader(),
Review Comment:
[P1] Keep external BlobView reads on a path that retains the dependency
context
With `blob-view-field` and native reading enabled, clearing this loader also
changes `_load_native_table` to use `Table.from_resolved_schema`.
`_native_blob_view_supported` currently accepts `loader is None`, so this does
not fall back as described in the earlier reply. The resolved-schema Rust table
has `rest_env=None`; its Data Evolution reader therefore skips BlobView
resolution. The native batch adapter does not resolve these values afterward.
A real external-table read returned `b'\x01WEIVBOLB...'` instead of the
referenced `b'actual'`, and retained a reference for a row whose resolved value
is NULL. Both direct and nested references showed this; Python reads and native
controls retaining the previous REST loader returned the expected payload/NULL.
Please make loaderless external tables that need BlobView resolution fall back
to Python, or carry their REST dependency context into the native table without
reintroducing REST snapshot management.
Verification detail: the local Rust 0.4.0 extension predates the current
scheduling arguments, so the focused native probe used a test-only adapter
omitting `parallelism`/`ordered`, with Python planning and Python-read fallback
forbidden. Current Rust source independently confirms the missing REST
environment and skipped resolution. The normal Python suites passed 241 tests;
the unadapted native matrix could not complete against this older extension.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]