[manuzhang]

What

Re-enables clang-tidy's misc-include-cleaner, scoped to standard-library headers only.

Why

#820 enabled the check unscoped and #708 removed it five days later because it demanded a direct include for every project symbol (iceberg::Result, iceberg::Status, ...) that an already-included project header provides, which blocked CI with long include lists. Since then nothing checks includes, and review of #1066 found regressions of this shape: history_table.cc drops <vector> but still uses std::vector, and the new metadata_scan.cc uses std::vector, std::shared_ptr, std::unique_ptr, std::string and std::make_unique without <vector>, <memory> or <string>. Both compile only because metadata_scan.h happens to include those headers.

How

misc-include-cleaner.IgnoreHeaders lists iceberg/.* and every third-party header family, so the only diagnostics left are:

• a std:: symbol (or int64_t, size_t, ...) used without a direct include of its standard header
• a standard header that is included but not used

With cpp-linter already running clang-tidy 22 on changed lines, this needs no new workflow or tooling.

Measured on main

Running the scoped check over every library and test source with the same clang-tidy version CI uses (22.1.8):

Category	Warnings	Files
std symbol used without its header	2271	303
std header included but unused	29	19
project / third-party header (ignored by this config)	2622	–

Unscoped, the check would emit all 4922; scoped, the project-header half disappears entirely.

The pre-existing standard-library gaps are not fixed here. Because cpp-linter runs with --lines-changed-only=true, they only surface on lines a PR touches, so the codebase converges incrementally and this PR stays a config-only change. A one-shot cleanup with clang-tidy --fix is possible as a follow-up if preferred.

Verified against #1066

Running the new config on that branch reports the missing <vector> in history_table.cc and the five missing standard headers in metadata_scan.cc, and nothing else in those files. One caveat: the std::vector use in history_table.cc is on a line #1066 did not change, so CI's changed-lines mode would report only the metadata_scan.cc findings. Catching include removals whose use sites are untouched would require whole-file linting of changed files, which is a separate decision.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

The .clang-tidy exclusions omit zlib and platform headers, allowing non-standard headers to trigger diagnostics despite the intended standard-library-only scope.

1 open issue

• Exclude zlib and platform headers from diagnostics · New

What changed in this PR

Re-enables misc-include-cleaner in the existing clang-tidy setup, aiming to enforce standard-library include hygiene incrementally without restoring project-header noise.

Changes:

• Enables missing- and unused-include diagnostics.
• Adds exclusions for Iceberg and third-party headers while leaving the changed-lines-only CI workflow unchanged.

File	Description
.clang-tidy	Enables include checking and configures header exclusions intended to limit diagnostics to standard-library headers.

_{Review effort: Balanced}