Extend ResultExportService unit test coverage

[mengw15]

Task Summary

ResultExportService (amber/src/main/scala/org/apache/texera/web/service/ResultExportService.scala) is at ~6% — 160 of 170 tracked lines are unhit. The existing ResultExportServiceSpec only covers parseOperators and validateExportRequest. The five streamDocumentAs* writers look storage-bound but are not: each takes a VirtualDocument[Tuple] parameter, and VirtualDocument is an ordinary abstract class whose only abstract members are getURI and clear() — every accessor defaults to throw new NotImplementedError. A small in-spec fake overriding getCount / get() / getRange / asInputStream() makes the whole export-encoding layer testable against a ByteArrayOutputStream, with no Iceberg, MinIO or DB.

Behavior to add

Extend amber/src/test/scala/org/apache/texera/web/service/ResultExportServiceSpec.scala. The service constructs cheaply — the spec already does new ResultExportService(WorkflowIdentity(1L), computingUnitId = 0). Reach the private writers with PrivateMethodTester (the idiom WorkflowExecutionsResourceSpec already uses). For the fake document, mirror MinimalDoc in common/workflow-core/src/test/scala/org/apache/texera/amber/core/storage/model/VirtualDocumentSpec.scala — override only what each test needs.

streamDocumentAsCSV (private; doc, outputStream, maybeHeaders: Option[List[String]])

• getCount == 0 → returns without writing a byte.
• maybeHeaders = None → header row is inferred from firstRow.getSchema.getAttributeNames, and that first row is also written to the body.
• maybeHeaders = Some(hdrs) → hdrs is written as the header and the iterator's first row is not consumed, so it still appears in the body. This asymmetry between the two branches is the off-by-one worth pinning.
• A document larger than Constants.CHUNK_SIZE (10) exercises the buffered chunk loop and the per-chunk flush().

streamCellData (private; out, request, operatorDocument)

• rowIndex >= getCount → WebApplicationException with message Invalid rowIndex (5). Total rows: 2.
• columnIndex >= selectedRow.getFields.length → WebApplicationException with the Invalid columnIndex message.
• Happy path writes exactly the UTF-8 bytes of the selected cell.

convertFieldToBytes (private) — three branches: an Array[Byte] passes through unchanged; a String encodes as UTF-8 (use a multi-byte character to pin the charset); anything else goes through toString.

streamDocumentAsHTML (private) — writes results.head.getField(0).toString as UTF-8. An empty document hits an unguarded .head and throws NoSuchElementException; assert that rather than leaving it undocumented.

streamDocumentAsParquetZip (private) — with a fake whose asInputStream() returns a ByteArrayInputStream, assert the bytes are copied through verbatim and the source stream is closed.

streamDocumentAsArrow (private) — runs entirely in-process on a RootAllocator; assert the Arrow stream round-trips the tuples back. Include only if it keeps the PR reviewable; it is the largest single block.

NonClosingOutputStream.close — asserts the wrapped stream is not closed.

Out of scope

getOperatorDocument, saveToDatasets, saveStreamToDataset and the exportSingleOperatorToDataset happy path need DocumentFactory.openDocument and a live file-service HTTP endpoint — leave them uncovered. Note fileServiceUploadOneFileToDatasetEndpoint is a lazy val defaulting to http://localhost:9092/..., so any test that reaches saveToDatasets will attempt a real socket connection.

generateFileName embeds LocalDateTime.now() in the name. If it is covered at all, assert only the stable parts — the ${workflowName}-op$operatorId-v$latestVersion- prefix, the parquet → .zip extension mapping, and that StringUtils.replaceEach strips / and \ from the workflow name. Never assert the full filename.

Task Type

• Refactor / Cleanup
• DevOps / Deployment / CI
• Testing / QA
• Documentation
• Performance
• Other