[lucasfang]

Purpose

Linked issue: close #488

The coverage report generated by test/generate_coverage.sh currently includes system headers (/usr/include/c++/...), third-party headers installed into the build tree (arrow_ep-install/, boost_ep-install/, etc.), third_party/ sources, and test sources (*_test.cpp, test helpers, mocks). This floods the directory view with tens of thousands of irrelevant files and dilutes the overall coverage numbers.

Add an lcov --remove step to focus the report on project implementation code, and enable AVRO, ORC, MOSAIC, LANCE, and VORTEX in the coverage build so optional formats are measured too.

Restrict build-tree exclusion to the actual build directory ($PWD/*). The previous */build*/* pattern also matched checkout ancestors such as /builds/, which could remove project implementation sources from the report. Pass --ignore-errors unused so overlapping or unmatched exclusion patterns do not abort filtering.

Anchor all remaining project exclusions to the repository root as well. Checkouts beneath test, testing, third_party, or mock-build must retain implementation sources. Exclude the known src/paimon/testing/ directory and explicit REST mock server files instead of matching mock anywhere in a path. Quote the initial directory change to support checkout paths containing spaces.

Tests

Validation for the latest follow-up commit 8208bd8c:

• bash -n test/generate_coverage.sh, pre-commit run --files test/generate_coverage.sh, and git diff --check — passed.
• Executed the exact filtering block against synthetic tracefiles for checkouts beneath normal, builds, test, testing, mock-build, third_party, and path with spaces. All seven scenarios retained implementation source/header records, including a production source within a mock-build directory, and removed build dependencies, system headers, third-party sources, tests, helpers, and known mocks. Implementation-only traces also passed with unmatched exclusions. genhtml --no-source generated HTML in all 14 cases.
• Reproduced the previous rules removing all implementation records beneath test, testing, mock-build, and third_party ancestors.
• Repeated all seven path scenarios with mixed and implementation-only tracefiles using LCOV 2.0-1 after installing its Perl dependencies in an isolated temporary directory. All 14 filtering and HTML-generation cases passed. Temporary source files were provided for LCOV 2.0's source-existence checks. Earlier checks also passed with the installed lcov (reports version 1.0.0).

Validation for follow-up commit dbb2278b:

• bash -n test/generate_coverage.sh — passed.
• pre-commit run --files test/generate_coverage.sh — passed; hooks not applicable to shell files were skipped.
• git diff --cached --check — passed before committing.
• Executed the script's exact lcov --remove block against synthetic tracefiles using the installed lcov (reports version 1.0.0). In a checkout under /builds/, implementation source/header records were retained while system, build dependency, third-party, test, helper, and mock records were removed. A tracefile containing only implementation records also completed successfully with unmatched exclusion patterns.

The original patch's previously reported internal-fork validation was a full build and 44/44 CTest targets passing, with filtered coverage of 93.1% lines and 92.3% functions. Those figures predate this follow-up; the complete coverage build and CTest suite were not rerun for this script-only change. Other lcov versions were not tested.

API and Format

No changes to public API, storage format, or protocol.

Documentation

Script comments explain the exclusions, build-directory scope, and handling of unused patterns. No new user-facing feature.

Generative AI tooling

Generated-by: Qoder
Generated-by: OpenAI Codex (GPT-6)

[SteNicholas]

@lucasfang, I left two comments for the coverage filtering step, reproduced with synthetic coverage traces using LCOV 2.0.

[SteNicholas]

@lucasfang, the two earlier findings are addressed. I left only one related path-filtering comment which has been reproduced with synthetic coverage traces using LCOV 2.0.

[SteNicholas]

LGTM.

[lxy-9602]

+1