All: Remove deprecated methods for 1.12.0 by dramaticlly · Pull Request #16449 · apache/iceberg

dramaticlly · 2026-05-20T18:02:11Z

Remove all methods, classes, and fields marked with "@deprecated will be removed in 1.12.0" across all modules, since 1.11.0 has been released.

Deleted Classes

SystemProperties → use SystemConfigs
PartitionStats → use PartitionStatistics interface / BasePartitionStatistics
DataReader → use PlannedDataReader
GenericAppenderFactory → use GenericFileWriterFactory
BaseFileWriterFactory → use RegistryBasedFileWriterFactory
S3SignRequest, S3SignResponse, S3SignRequestParser, S3SignResponseParser, S3ObjectMapper → use RemoteSign* equivalents

Notable Non-Mechanical Changes

Beyond deleting deprecated code, the following adjustments were needed. Items marked ⚠️ reviewer attention depart from a strict deprecation removal and warrant a closer look:

⚠️ PositionDelete.set(path, pos, row) and row() — Kept without @Deprecated. The public deprecation notice said position deletes with row data are no longer supported, but internal code (BaseTaskWriter, BasePositionDeltaWriter, RewriteTablePathUtil, Avro) still relies on these methods. Reviewer: confirm the methods are safe to retain as package-internal helpers, or whether a follow-up should narrow visibility / push removal to 1.13.0.
⚠️ HadoopFileIO.serializeConfWith() — Kept without @Deprecated because the HadoopConfigurable interface still declares the method. Also inlined the constructor logic since the removed constructor was the delegation target. Reviewer: the cleaner path is to deprecate the interface method (and the ResolvingFileIO / SerializableFileIOWithSize × 3 implementations) for 1.13.0 removal, then drop the SerializationUtil instanceof HadoopConfigurable branch and the iceberg.mr.config.serialization.disabled opt-in — those are the only remaining consumers after Core: Move Hadoop conf serialization into SerializableConfiguration #15583 made SerializableConfiguration itself SerializableSupplier-compatible. Happy to land this as a follow-up.
BaseParquetReaders.createStructReader(List, StructType, Integer) — Changed from a concrete method (that delegated to the removed 2-param version) to abstract. Both existing subclasses (InternalReader, GenericParquetReaders) already override it.
⚠️ S3V4RestSignerClient — Removed deprecated property fallback logic from baseSignerUri(), endpoint(), and check(). The legacy properties (s3.signer.uri, s3.signer.endpoint) are no longer supported, and endpoint() now requires RESTCatalogProperties.SIGNER_ENDPOINT (previously defaulted to "v1/aws/s3/sign"). Reviewer: this is a stricter behavior change than a pure removal — please confirm downstream catalog adapters / integration tests are aware they must now supply RESTCatalogProperties.SIGNER_ENDPOINT explicitly.

AI Disclosure

Model: Claude Opus 4.7 (1M context)
Platform/Tool: Opus
Human Oversight: reviewed
Prompt Summary: provide class and method deprecation scheduled for upcoming 1.12.0 release

nastra · 2026-05-22T10:20:20Z

-    // TODO change to required in 1.12.0
-    if (!properties().containsKey(S3_SIGNER_ENDPOINT)
-        && !properties().containsKey(RESTCatalogProperties.SIGNER_ENDPOINT)) {
-      LOG.warn(


this shouldn't be removed. Instead, this should go through a Preconditions.checkArgument check

yeah I think RESTCatalogProperties.SIGNER_ENDPOINT is replaced with RESTCatalogProperties.SIGNER_URI and we can rely on LINE 206 to ensure its existence. Sorry it's a bit messy right now

@adutra can you help check the deprecation align with your original goal in #15451.

Removes S3_SIGNER_URI / S3_SIGNER_ENDPOINT / S3_SIGNER_DEFAULT_ENDPOINT

RESTCatalogProperties.SIGNER_ENDPOINT is now required

dramaticlly · 2026-05-26T06:37:22Z

@nastra do you want to take another look as now CI is green. I leave the PositionDelete.set(path, pos, row) and row() as is given there's nontrivial amount of caller still use the Deprecated method, plan to handle in a follow up PR

…NSAFE branch The prior 1.11 logic in `Parquet.ReadBuilder` had three branches when picking the name mapping for a vectorized read: 1. caller-provided nameMapping → use it 2. `NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED=true` → use null (unsafe fallback to position-based field resolution) 3. default → use `NameMapping.empty()` (safe) PR apache#16449 removed the deprecated NETFLIX_UNSAFE branch (case 2) but collapsed the default to `null` rather than `NameMapping.empty()`, making unsafe position-based resolution the new default. That contradicts the PR description ("always use NameMapping.empty() (the safe default)") and silently regresses readers that don't supply an explicit mapping. Restore the safe default by using `NameMapping.empty()` when no mapping is provided. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`build.gradle` line 145 sets `oldVersion = "1.11.0"` — revapi compares the current code against the 1.11.0 released baseline. Section keys in `.palantir/revapi.yml` correspond to that baseline, so new breaks accepted during the 1.12.0 dev cycle belong under `"1.11.0"`, not under a not-yet-released `"1.12.0"` section. Evidence in-file: the `"1.10.0"` section already contains entries with justifications like "Removing deprecated code for 1.11.0" — entries added during the 1.11.0 dev cycle when `oldVersion` was `"1.10.0"`. Same shape applies now. This commit: - Renames the section header `"1.12.0"` -> `"1.11.0"`, merging the new entries from PR apache#16449 into the existing 1.11.0 section that was already on upstream/main (the partition-stats entries from PR apache#14998). - Restores the original justifications on the 17 partition-stats entries that were rewritten by an earlier fixup commit (`fb5ddae9b2 Core: Move partition stats revapi entries to 1.12.0`), back to "Removed deprecated functionality for partition stats". - Drops a duplicate `class org.apache.iceberg.PartitionStats` entry that the earlier fixup commit had introduced. Net effect: the cumulative PR diff against upstream/main is now additive only (new entries appended to the existing 1.11.0 section, no churn on existing entries). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

manuzhang · 2026-07-20T03:53:15Z

@dramaticlly are you still working on this?

…NSAFE branch The prior 1.11 logic in `Parquet.ReadBuilder` had three branches when picking the name mapping for a vectorized read: 1. caller-provided nameMapping → use it 2. `NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED=true` → use null (unsafe fallback to position-based field resolution) 3. default → use `NameMapping.empty()` (safe) PR apache#16449 removed the deprecated NETFLIX_UNSAFE branch (case 2) but collapsed the default to `null` rather than `NameMapping.empty()`, making unsafe position-based resolution the new default. That contradicts the PR description ("always use NameMapping.empty() (the safe default)") and silently regresses readers that don't supply an explicit mapping. Restore the safe default by using `NameMapping.empty()` when no mapping is provided. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`build.gradle` line 145 sets `oldVersion = "1.11.0"` — revapi compares the current code against the 1.11.0 released baseline. Section keys in `.palantir/revapi.yml` correspond to that baseline, so new breaks accepted during the 1.12.0 dev cycle belong under `"1.11.0"`, not under a not-yet-released `"1.12.0"` section. Evidence in-file: the `"1.10.0"` section already contains entries with justifications like "Removing deprecated code for 1.11.0" — entries added during the 1.11.0 dev cycle when `oldVersion` was `"1.10.0"`. Same shape applies now. This commit: - Renames the section header `"1.12.0"` -> `"1.11.0"`, merging the new entries from PR apache#16449 into the existing 1.11.0 section that was already on upstream/main (the partition-stats entries from PR apache#14998). - Restores the original justifications on the 17 partition-stats entries that were rewritten by an earlier fixup commit (`fb5ddae9b2 Core: Move partition stats revapi entries to 1.12.0`), back to "Removed deprecated functionality for partition stats". - Drops a duplicate `class org.apache.iceberg.PartitionStats` entry that the earlier fixup commit had introduced. Net effect: the cumulative PR diff against upstream/main is now additive only (new entries appended to the existing 1.11.0 section, no churn on existing entries). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

…DeleteRowSchema - TestS3FileIOProperties: Add required signer.endpoint property since the deprecated default was removed - TestSparkWriterMetrics: Override checkRowStatistics and checkNotExistingRowStatistics to match behavior without positionDeleteRowSchema (consistent with Flink tests) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

BaseScan.io() and PositionDelete.set(path, pos, row)/row() still have callers and cannot be removed yet. Restore their deprecation annotations to preserve the original removal intent. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…io() The method had only 4 internal callers (DataScan, BaseDistributedDataScan) which are trivially replaced with table().io(). Also moves revapi entries to the 1.12.0 section with unified justification. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

…ification All deprecated API removals for this release should be under the "1.12.0" version key with justification "Removing deprecated API scheduled for removal in 1.12.0", sorted alphabetically by class/method name. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

The `specsById` map carried on `BaseScanTaskResponse` and its builder is parser context, not part of the REST spec for scan-planning response models. It was deprecated in 1.11.0 (apache#14485) with a note that visibility would be reduced in 1.12.0. Demote `specsById()` getter and `withSpecsById()` setter (plus the builder's `deleteFiles()` getter) from public to protected so the parsers and subclasses can still use them while keeping the field out of the public API. To preserve cross-package construction paths used by `CatalogHandlers`, `RESTServerCatalogAdapter`, and tests in `org.apache.iceberg.rest`: - Add `builder(Map<Integer, PartitionSpec> specsById)` factory on each of `PlanTableScanResponse`, `FetchPlanningResultResponse`, and `FetchScanTasksResponse` for initial server-side construction. - Add `toBuilder()` on each response for copy-with-modification patterns (used when adapters rebuild a response to inject credentials or rewrite status). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…NSAFE branch The prior 1.11 logic in `Parquet.ReadBuilder` had three branches when picking the name mapping for a vectorized read: 1. caller-provided nameMapping → use it 2. `NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED=true` → use null (unsafe fallback to position-based field resolution) 3. default → use `NameMapping.empty()` (safe) PR apache#16449 removed the deprecated NETFLIX_UNSAFE branch (case 2) but collapsed the default to `null` rather than `NameMapping.empty()`, making unsafe position-based resolution the new default. That contradicts the PR description ("always use NameMapping.empty() (the safe default)") and silently regresses readers that don't supply an explicit mapping. Restore the safe default by using `NameMapping.empty()` when no mapping is provided. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`build.gradle` line 145 sets `oldVersion = "1.11.0"` — revapi compares the current code against the 1.11.0 released baseline. Section keys in `.palantir/revapi.yml` correspond to that baseline, so new breaks accepted during the 1.12.0 dev cycle belong under `"1.11.0"`, not under a not-yet-released `"1.12.0"` section. Evidence in-file: the `"1.10.0"` section already contains entries with justifications like "Removing deprecated code for 1.11.0" — entries added during the 1.11.0 dev cycle when `oldVersion` was `"1.10.0"`. Same shape applies now. This commit: - Renames the section header `"1.12.0"` -> `"1.11.0"`, merging the new entries from PR apache#16449 into the existing 1.11.0 section that was already on upstream/main (the partition-stats entries from PR apache#14998). - Restores the original justifications on the 17 partition-stats entries that were rewritten by an earlier fixup commit (`fb5ddae9b2 Core: Move partition stats revapi entries to 1.12.0`), back to "Removed deprecated functionality for partition stats". - Drops a duplicate `class org.apache.iceberg.PartitionStats` entry that the earlier fixup commit had introduced. Net effect: the cumulative PR diff against upstream/main is now additive only (new entries appended to the existing 1.11.0 section, no churn on existing entries). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

After restoring `NameMapping.empty()` as the default in `Parquet.ReadBuilder` (commit 8881516), two data-module tests fail because they wrote parquet files via raw `AvroParquetWriter` (no Iceberg field IDs in the file metadata) and read them back without providing a `NameMapping`, implicitly relying on the unsafe positional-fallback ID assignment that the `NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED` config previously enabled by default. The deprecation message on that config explicitly said "Fallback ID assignment in Parquet is UNSAFE and will be removed in 1.12.0. Use name mapping instead." Honor that guidance: - `TestGenericData.testTwoLevelList` and `TestParquetEncryptionWithWriteSupport.testTwoLevelList` now pass an explicit `withNameMapping(MappingUtil.create(schema))` to the reader. - Add `TestGenericData.testReadWithoutFieldIdsOrNameMappingReturnsNullFields` to assert the new strict default behavior: when a parquet file has no field IDs and no `NameMapping` is supplied, projected fields read as null (rather than being bound by position as before). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…itional-fallback Same root cause as the data-module migrations in ac9ee16: the test writes a parquet file via raw AvroParquetWriter (no Iceberg field IDs) and then reads it back without supplying a NameMapping. Pre-1.12 this worked via the NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED positional fallback (defaulted to true on upstream/main); after this PR's Parquet ReadBuilder restores NameMapping.empty() as the default, projected fields no longer bind by position and the assertion NPEs. Migrate all three flink versions (v1.20, v2.0, v2.1) to supply `withNameMapping(MappingUtil.create(schema))`, matching the strict default and aligning with the deprecation message's guidance to "Use name mapping instead". Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

After restoring `NameMapping.empty()` as the safe default in `Parquet.ReadBuilder` (commit 8881516), Spark tests that import external Parquet files via `SparkTableUtil.importSparkTable` and then read them through Iceberg fail field resolution: the imported files have no Iceberg field IDs and no `schema.name-mapping.default` property is auto-set, so the strict empty mapping returns null for every projected field. Mirror the established pattern from `AddFilesProcedure.ensureNameMappingPresent` and `BaseTableCreationSparkAction`: when the target Iceberg table does not already have `schema.name-mapping.default` set, auto-derive a name mapping from the target schema before importing. Subsequent reads via Iceberg pick up the property automatically through `BaseReader`. Applied identically to v3.5, v4.0, and v4.1. This honors the deprecation guidance Russell Spitzer documented when deprecating the NETFLIX_UNSAFE positional fallback ("Use name mapping instead") and brings `importSparkTable` in line with `AddFilesProcedure`, which has been doing this for some time. Verified on Spark 4.1 via the full `org.apache.iceberg.spark.source.*` test package (`-DtestParallelism=8`); previously-failing `TestIcebergSourceHadoopTables.testTableWithInt96Timestamp` and `TestIdentityPartitionData.testProjections` now pass with no collateral failures elsewhere. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…ate TestSparkParquetReader fixtures After restoring `NameMapping.empty()` as the safe default in `Parquet.ReadBuilder` (commit 8881516), `IcebergGenerics.read(table)` started failing field resolution on tables whose data files lack Iceberg field IDs. Pre-1.12 the unsafe positional-fallback path masked this; now the strict default refuses to bind by position. `GenericReader` (data module) never consulted `schema.name-mapping.default` even though every Iceberg table that reads non-Iceberg-written files is expected to carry it. Spark's `BaseReader` already reads the property (`BaseReader.java:100`); bring `GenericReader` in line with that pattern so `IcebergGenerics.read(table)` auto-applies the mapping when present. Also migrate `TestSparkParquetReader.testInt96TimestampProducedBySparkIsReadCorrectly` (v3.5 / v4.0 / v4.1) — the test creates a `HadoopTables` table directly (bypassing `SparkTableUtil.importSparkTable`, so the auto-set in commit 9ebb273 doesn't reach it) and reads files written by raw Spark `ParquetWriteSupport` without Iceberg field IDs: - `rowsFromFile`: pass `.withNameMapping(MappingUtil.create(schema))` on the direct `Parquet.read(...)` call. - `tableFromInputFile`: set `schema.name-mapping.default` at table creation time so the `GenericReader` change above picks it up automatically. Verified: - `:iceberg-data:test` clean - Targeted `TestSparkParquetReader.testInt96TimestampProducedBySparkIsReadCorrectly` passes on v3.5, v4.0, v4.1 - `:iceberg-flink:iceberg-flink-2.1:test` full sweep clean (30m, `-DtestParallelism=8`) - Spotless / checkstyle / revapi clean across all versions Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Migrate the vectorized dictionary-encoded parquet read test off unsafe positional-fallback ID assignment by attaching a NameMapping created from the schema to the Parquet read builder. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

@OverRide

Upstream commit 773907d ("Core: Add id tracking to MetricsConfig") removed checkNotExistingRowStatistics() from the abstract TestWriterMetrics parent and removed MetricsConfig.forPositionDelete(Table). After rebasing onto that commit, our branch had orphaned overrides and a divergent revapi justification. - Remove the now-orphaned @OverRide checkNotExistingRowStatistics from TestSparkWriterMetrics in spark v3.5/v4.0/v4.1 (the parent method no longer exists); the checkRowStatistics override remains valid. - Align revapi.yml justifications with upstream: use "Deprecated for removal in 1.12.0" for MetricsConfig::forPositionDelete(Table) and the uniform partition-stats wording. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- AWS: Rename legacySignerProperties to signerUriResolution (no legacy properties remain) and add signerEndpointIsRequired coverage for the now-required signer.endpoint precondition. - Core: Replace hardcoded "%1F"/"%2E" separator literals in TestRESTUtil and TestResourcePaths with the canonical NAMESPACE_SEPARATOR_URLENCODED_UTF_8 constants so the tests stay coupled to source. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

dramaticlly · 2026-07-24T20:44:06Z

-    // TODO change to required in 1.12.0
-    if (!properties().containsKey(S3_SIGNER_ENDPOINT)
-        && !properties().containsKey(RESTCatalogProperties.SIGNER_ENDPOINT)) {
-      LOG.warn(


@adutra can you help check the deprecation align with your original goal in #15451.

Removes S3_SIGNER_URI / S3_SIGNER_ENDPOINT / S3_SIGNER_DEFAULT_ENDPOINT

RESTCatalogProperties.SIGNER_ENDPOINT is now required

dramaticlly · 2026-07-24T23:00:47Z

-    // Fallback to the signature without fieldId if not overridden
-    return createStructReader(fieldReaders, structType);
-  }
+  protected abstract ParquetValueReader<T> createStructReader(


@RussellSpitzer I think with createStructReader(fieldReaders, structType) removed and all subclass override the method, this can now become abstract method, do you think this is safe to apply?

dramaticlly · 2026-07-24T23:02:19Z

-  public static ManifestReader<DataFile> read(ManifestFile manifest, FileIO io) {
-    return read(manifest, io, null);


@RussellSpitzer I think the public static read method might be used a lot outside iceberg package, want to double check this is ok to remove with 1 release between deprecation and removal

dramaticlly · 2026-07-24T23:05:32Z

-   * @deprecated will be removed in 1.12.0; use name mapping instead
-   */
-  @Deprecated
-  public static final ConfigEntry<Boolean> NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED =


@rdblue I want to double check with you on this. More in #16449 (comment), I end up changing few spark and Flink test for using NameMapping.empty() instead of null (unsafe positional ID assignment).

dramaticlly · 2026-07-24T23:06:21Z

-   *     TableIdentifier, SnapshotMode)} instead.
-   */
-  @Deprecated
-  public static LoadTableResponse loadTable(Catalog catalog, TableIdentifier ident) {


@gaborkaszab I think this 2 arguments public static method might be used a lot outside iceberg package, want to double check this is ok to remove with 1 release between deprecation and removal.

dramaticlly · 2026-07-24T23:11:16Z

-  /**
-   * @deprecated since 1.11.0, will be removed in 1.12.0.
-   */
-  @Deprecated


@nastra want to check with you that on not remove this for now, as the HadoopFileIO today still implement the required method defined in HadoopConfigurable interface. Can follow up with better removal cycle in 1.13 if needed.

dramaticlly · 2026-07-24T23:13:34Z


-  @SuppressWarnings("deprecation")
-  public static Stream<Arguments> legacySignerProperties() {
+  public static Stream<Arguments> signerUriResolution() {


@nastra I renamed method source and UT above as after this PR we no longer have legacySignerProperties

dramaticlly · 2026-07-24T23:16:44Z

  }

-  /**
-   * @deprecated since 1.11.0, will be removed in 1.12.0; use {@link BaseScan#fileIO()} instead.


For existing scan which use io to read manifests, I end up using table().io() instead of fileIO().get()` as it reads better

dramaticlly · 2026-07-24T23:19:53Z

-     */
-    @Deprecated
-    public B withSpecsById(Map<Integer, PartitionSpec> specs) {
+    protected B withSpecsById(Map<Integer, PartitionSpec> specs) {


@singhpk234 can you help check the visibility change from public to protected align with your original goal, also end up providing sanctioned entry point of witSpecById in builder for BaseScanTaskResponse,FetchPlanningResultResponse,FetchScanTasksResponse,PlanTableScanResponse,

u70b3

Thanks for working through this broad cleanup. I am new to the Iceberg community, so please correct me if I have missed an intended calling constraint. I ran focused reproductions against the current PR head and found two cases that seem worth checking.

u70b3 · 2026-07-30T13:49:50Z

          "Table %s does not exist", sourceTableIdentWithDB);
    }

+    ensureNameMappingPresent(targetTable);


Could we also ensure the name mapping in the deepest retained importSparkPartitions overload (or otherwise ensure that every public overload reaches this initialization)? I may be missing an intended precondition, but I reproduced this on the current PR head by adapting the existing TestIdentityPartitionData setup to call public getPartitions + importSparkPartitions directly. Both Parquet parameterizations failed: the partition columns remained correct, but every id and message value was read as null. Adding ensureNameMappingPresent(targetTable) to the deepest overload made all 5 testFullProjection parameterizations pass. I ran: ./gradlew :iceberg-spark:iceberg-spark-4.1_2.13:test --tests "org.apache.iceberg.spark.source.TestIdentityPartitionData.testFullProjection".

Thanks @u70b3 , I think after some deep dive we decided to keep NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED for a few more releases given the silent behavior change of imported parquet files in #17444

u70b3 · 2026-07-30T13:49:50Z

@@ -93,20 +89,7 @@ public B withFileScanTasks(List<FileScanTask> tasks) {
      return self();


Could withFileScanTasks(null) clear the builder's derived deleteFiles as well? I may be misunderstanding the builder contract, but I reproduced this by creating a completed response whose task has a delete file, then calling response.toBuilder().withPlanStatus(FAILED).withFileScanTasks(null).build(). It throws IllegalArgumentException: deleteFiles should only be returned with fileScanTasks that reference them because the copied delete files remain. Setting deleteFiles to null when tasks are cleared made the focused test pass. I ran: ./gradlew :iceberg-core:test --tests "org.apache.iceberg.rest.responses.TestPlanTableScanResponseParser.toBuilderClearsDeleteFilesWhenClearingFileScanTasks".

I think this is a good callout, applied suggestion to remove deleteFiles in 7024529

@deprecated

Reverts the removal of SystemConfigs.NETFLIX_UNSAFE_PARQUET_ID_FALLBACK_ENABLED, the Parquet.ReadBuilder change it guarded, and everything that followed from it: the GenericReader name-mapping lookup, the SparkTableUtil deprecation removals and name-mapping auto-set, and the migrated Parquet reader tests in data, flink and spark. That config defaulted to true, so removing it changes how every Iceberg deployment reads Parquet files without field IDs, and the failure mode is silent nulls rather than an error. It needs a dedicated review thread on the migration path instead of riding along in a broad deprecation sweep. SparkTableUtil moves with it in full, including its own "@deprecated since 1.11.0, will be removed in 1.12.0" members, so that exactly one PR owns the file and the two do not conflict. This PR keeps every other 1.12.0 removal.

BaseScanTaskResponse.Builder.withFileScanTasks derives deleteFiles from the tasks, but only on the non-null branch, so clearing the tasks left the previously derived delete files behind. Combined with the removal of withDeleteFiles in this PR, there was no longer any way to clear them, and response.toBuilder().withPlanStatus(FAILED).withFileScanTasks(null).build() failed validation with "deleteFiles should only be returned with fileScanTasks that reference them". Derives on both branches so clearing the tasks clears the derived delete files, and reads the tasks parameter rather than the field, which is what the guard always meant. Reported-by: u70b3

singhpk234

Thanks @dramaticlly, just took a pass on rest changes in this cycles, added comments inline

singhpk234 · 2026-07-31T18:10:03Z

+      this.deleteFiles =
+          tasks == null
+              ? null
+              : DeleteFileSet.of(
+                  () -> tasks.stream().flatMap(task -> task.deletes().stream()).iterator());


is this required ?

the logic is revised slightly so that now we can clean this.deleteFiles when passing null tasks. This is needed as we are removing the seter of withDeleteFiles in the same PR, so need a way to reset the internal deleteFiles if they are not longer applicable. Before the change it will only set this.deleteFiles if we have a valid task but not be able to clear

singhpk234 · 2026-07-31T18:10:42Z

+  /**
+   * Returns a builder pre-populated with this response's fields, suitable for producing a copy with
+   * one or more fields modified.
+   */
+  public Builder toBuilder() {
+    return new Builder()
+        .withPlanStatus(planStatus)
+        .withErrorResponse(errorResponse)
+        .withPlanTasks(planTasks())
+        .withFileScanTasks(fileScanTasks())
+        .withCredentials(credentials())
+        .withSpecsById(specsById());
+  }


is this required ?

I added this to provide a way for class outside the package to create the builder correctly as now the specById is reduced to protected scope. Current user are org.apache.iceberg.rest.RESTServerCatalogAdapter and org.apache.iceberg.rest.TestRESTScanPlanning

github-actions Bot added spark parquet core data flink ORC AWS GCP KAFKACONNECT labels May 20, 2026

nastra reviewed May 22, 2026

View reviewed changes

dramaticlly force-pushed the 1.12deprecation branch from cadfa6b to d62346b Compare May 23, 2026 00:04

github-actions Bot removed the ORC label May 23, 2026

dramaticlly force-pushed the 1.12deprecation branch 3 times, most recently from aa6430d to 0436357 Compare May 25, 2026 17:08

dramaticlly marked this pull request as ready for review May 26, 2026 06:35

gaborkaszab mentioned this pull request May 26, 2026

Core: Remove deprecated APIs for 1.12.0 #16452

Draft

5 tasks

nssalian added this to the Iceberg 1.12.0 milestone Jun 13, 2026

manuzhang mentioned this pull request Jun 20, 2026

Remove deprecated methods for 1.12.0 #16445

Open

manuzhang requested a review from nastra June 20, 2026 12:25

github-actions Bot added the OPENAPI label Jun 25, 2026

dramaticlly force-pushed the 1.12deprecation branch from 73bda60 to 13b5244 Compare June 25, 2026 07:31

dramaticlly force-pushed the 1.12deprecation branch from 13b5244 to 8881516 Compare June 25, 2026 07:32

nastra requested review from pvary and singhpk234 June 25, 2026 09:24

dramaticlly mentioned this pull request Jul 1, 2026

Core: Add id tracking to MetricsConfig #17022

Merged

dramaticlly force-pushed the 1.12deprecation branch from ea434b8 to a198aac Compare July 22, 2026 22:38

dramaticlly and others added 15 commits July 23, 2026 09:39

All: Remove deprecated methods for 1.12.0

c1669e7

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

AWS: Fix testApplySignerConfiguration missing required signer.endpoint

4d9ecee

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

dramaticlly force-pushed the 1.12deprecation branch from b9e217d to 3729278 Compare July 23, 2026 20:20

dramaticlly commented Jul 24, 2026

View reviewed changes

pvary mentioned this pull request Jul 30, 2026

Core, Data: Validate equality delete field IDs #17028

Open

u70b3 reviewed Jul 30, 2026

View reviewed changes

dramaticlly added 2 commits July 30, 2026 20:38

singhpk234 reviewed Jul 31, 2026

View reviewed changes

dramaticlly requested a review from singhpk234 July 31, 2026 21:38

manuzhang mentioned this pull request Aug 5, 2026

deprecation for removal in 1.12.0 #17465

Open

		public static ManifestReader<DataFile> read(ManifestFile manifest, FileIO io) {
		return read(manifest, io, null);

		@@ -93,20 +89,7 @@ public B withFileScanTasks(List<FileScanTask> tasks) {
		return self();

Uh oh!

Conversation

dramaticlly commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Deleted Classes

Notable Non-Mechanical Changes

AI Disclosure

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dramaticlly commented May 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

manuzhang commented Jul 20, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

u70b3 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

singhpk234 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

dramaticlly commented May 20, 2026 •

edited

Loading

dramaticlly commented May 26, 2026 •

edited

Loading