Antalya 25.8 Backport of #96118 - Fix Gorilla codec decompression when data_bytes_size exceeds data size

[mkmkme]

25.8 Stable Backport of ClickHouse#96118 - Fix Gorilla codec decompression when data_bytes_size exceeds data size

Same as #2021, but for Antalya 25.8

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Fix a regression with the Gorilla codec when an explicitly specified size does not correspond to the data type size, and the buffer size is too small. In previous versions, it threw an exception on decompression. Closes ClickHouse#78253 (ClickHouse#96118 by @alexey-milovidov)

Documentation entry for user-facing changes

...

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Performance tests
• Aarch64 tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• All Regression
• Disable CI Cache

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• OAuth (5m)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

[github-actions]

Workflow [PR], commit [56df330]

[mkmkme]

@blau-ai

[blau-ai]

CI triage

Verdict: 2 red checks (+1 aggregate), 0 PR-caused. One is an infra build timeout, the other is an unrelated flaky test. The actual fix is verified — this PR's own new test passed.

The change here is 5 lines in src/Compression/CompressionCodecGorilla.cpp plus a new stateless test. Nothing it touches has any connection to the two failures below.

---

1. Build (amd_release) — infra timeout, not PR-caused

• Terminated with exit code [-15] (SIGTERM) after 2h3m18s.
• The build reached [15528/15531] Linking before it was killed — i.e. it finished compiling and was in the final ThinLTO link stage (the slowest part of a release build) when it hit the job time limit. No compile/link error was emitted.
• A 5-line change in a single translation unit cannot meaningfully affect release build time, and there is no error in the log — this is a runner timeout on the ThinLTO link.
• Next step: re-run the job. With a warm sccache it should finish the link inside the limit.

2. Stateless tests (amd_binary, ParallelReplicas, s3 storage, parallel) — flaky, not PR-caused

Report: Failed: 1, Passed: 7125, Skipped: 737, Broken: 2. The one real failure:

01710_projection_additional_filters: [ FAIL ]
Reason: result differs with reference:
@@ -1,2 +1,2 @@
 0	0	0
-3
+23

• This is a projections + additional-filters test running under parallel replicas with heavily randomized fuzzer settings (--parallel_replicas_local_plan 1, --optimize_aggregation_in_order 1, etc.). It has nothing to do with the Gorilla codec.
• Decisive evidence that the fix is fine: this PR's own new test passed —
  03823_gorilla_codec_mismatched_size: [ OK ] 0.93 sec.
• The 2 Broken entries are 03441_deltalake_clickhouse_virtual_columns / ..._public_datasets, which depend on external datasets (infra, not counted as failures).
• Next step: re-run. If 01710_projection_additional_filters keeps failing on re-run, it's a pre-existing flake in the ParallelReplicas matrix worth a separate issue — but it does not gate this backport.

3. PR (aggregate) — rolls up the two checks above; no independent error.

---

Bottom line: the backport is healthy. The codec fix behaves correctly (its dedicated test passes), and both red checks are infra/flake. Re-running Build (amd_release) and the ParallelReplicas stateless job should clear CI. No code change needed.