Repository navigation
GH-390: Refresh vector lookup when synchronizing the schema - #1389
Merged
jbonofre merged 1 commit intoNov 15, 2026
Merged
Conversation
hutiefang76
requested review from
jbonofre,
laurentgo,
lidavidm and
wgtmac
as code owners
October 24, 2026 08:37
Promoted child types change Field equality, leaving the lookup cache keyed by the old schema. Rebuild the cache when the schema changes so Field-based lookups and JSON serialization resolve the current vectors. Generated-by: Codex and DeepSeek
jbonofre
force-pushed
the
frank/arrow390-schema-sync-20261024
branch
from
November 14, 2026 07:55
17c93ce to
3afa607
Compare
jbonofre
added a commit
to jbonofre/arrow-java
that referenced
this pull request
Nov 15, 2026
… in VectorSchemaRoot JsonFileWriter and JsonFileReader look each vector up by position, so columns with equal Fields no longer share one vector. The Field lookup keeps the first of such columns, as getVector(String) documents, and syncSchema() keeps the custom metadata of the schema. Also addresses review comments on apache#1389: the constructor and syncSchema() share the code that builds the lookup, and the JSON round trip after a schema sync moved to TestJSONFile. Closes apache#1473.
jbonofre
added a commit
that referenced
this pull request
Nov 15, 2026
…ctorSchemaRoot (#1474) ## What's Changed Follow-up to #1389, which this PR is based on: until #1389 is merged, only the last commit is new. - `JsonFileWriter` and `JsonFileReader` look each vector up by position instead of through `getVector(Field)`, so columns with equal Fields (such as duplicate names with the same type) no longer share one vector. A JSON round trip of `x: [1], x: [2]` read back `x: [null], x: [2]` before. - The Field lookup of `VectorSchemaRoot` keeps the first of the columns with equal Fields, as `getVector(String)` documents, instead of the last. - `syncSchema()` keeps the custom metadata of the schema. - Review comments on #1389: the constructor and `syncSchema()` share the code that builds the lookup, the comments say the old keys no longer match because the promoted Field is structurally different, an assertion that could not fail is dropped, and the JSON round trip after a schema sync moved to `TestJSONFile`. `TestJSONFile` covers a JSON round trip of duplicate columns through both `read()` and `read(VectorSchemaRoot)`, and `TestVectorSchemaRoot` covers the lookup of duplicate columns and the metadata after `syncSchema()`. Closes #1473.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What's Changed
After a child type is promoted,
syncSchema()updates the schema but leaves the vector lookup keyed by the oldFieldvalues. Looking up a vector using the new schema returns null, andJsonFileWriterthen throws a NullPointerException. Refresh the Field-keyed lookup in the schema-change branch, preserving vector identity and ordering.Add regressions using two rows of promoted nested-list data: direct lookup verifies the new Field, removal of the stale key, unchanged values and row count, and a second synchronization that makes no change. A real JSON write/read round trip verifies the schema and both nested row values. The unchanged implementation fails the lookup assertion and throws in the writer.
Validation on JDK 21:
mvn -B -ntp -Dmaven.gitcommitid.nativegit=true -pl vector -am verifypassed, including dependency modules.TestVectorSchemaRootcases passed in each execution without skips.Closes #390.
Maintainer note: the CI workflows are
action_required, and my account cannot add repository labels. Please addbug-fixand approve the runs when appropriate. Remote CI and maintainer review remain pending.AI-generated PR description. Investigation, implementation and automated review assisted by Codex and DeepSeek.