hutiefang76 opened a new pull request, #12962:
URL: https://github.com/apache/seatunnel/pull/12962
### Purpose of this pull request
Yes, I am contributing this fix to Apache SeaTunnel.
Fixes #12961.
The JSON ARRAY and MAP converters currently iterate any JSON node without
checking its container type. As a result, an ARRAY field containing a string,
number or boolean silently becomes an empty array; an object becomes an array
of nulls. A MAP field containing a scalar or an array silently becomes an empty
map. This happens even when parse errors are not ignored and can silently
discard source data.
Validate the node shape before iterating it. ARRAY requires an array and MAP
requires an object. The resulting parse exception names the field and expected
container type and follows the existing error-handling wrappers. No dependency
or configuration option is added.
### Does this PR introduce _any_ user-facing change?
Yes. Malformed container values now fail deserialization in strict mode.
With `ignore-parse-errors = true`, only the malformed value becomes null. For a
malformed nested value, valid siblings and the enclosing collection remain
intact.
For example, with `arr` declared as ARRAY<INT>, `{"id":7,"arr":42}`
previously produced an empty array. It now reports a parse error, or yields
`id=7, arr=null` when errors are ignored. Valid empty containers, null/missing
fields, element and key conversions, and positional ROW arrays retain their
existing behavior.
English and Chinese compatibility notes explain the change and migration
options: supply the declared JSON container type, or enable ignored parse
errors to obtain null for malformed values.
### How was this patch tested?
Locally on macOS ARM64 with JDK 17.0.18, against dev commit
`6b63aca173eb8c056db0b78a87539aec3e4ad1a5`:
- Before changing production code, the public-schema regression run had 31
tests with 4 failing new test methods, covering 16 malformed input/mode
assertions. Existing and valid-input controls passed.
- Added seven tests in the existing schema test class. They cover
strict/ignored-error ARRAY and MAP inputs through both String and byte entry
points, field/type error context, valid empty/nonempty/nullable containers,
numeric-string coercion, and nested-value containment.
- `./mvnw -B -pl seatunnel-formats/seatunnel-format-json -am package`: 566
tests passed, including all 62 JSON-format tests, with no failures, errors or
skips.
- `./mvnw spotless:apply`: passed; a second format run left the patch
unchanged.
- `./mvnw -q -DskipTests verify`: passed (tests deliberately skipped in this
build check).
These are local Java format-module results; upstream CI remains pending.
AI assistance: Codex and DeepSeek assisted the investigation, patch, tests
and review. The submitted diff and actual test reports were checked before
submission.
### Check list
* [x] No new Jar binary package or dependency is added.
* [x] User-facing behavior and migration guidance are documented in English
and Chinese.
* [x] Both `incompatible-changes.md` files are updated.
* [x] Connector registration changes are not applicable to this format-only
patch.
--
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]
