Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

Thanks, this is a real bug: ares_dns_write() wrote a cut-down length for record data over 65535 bytes, which breaks the framing of every record after it. I confirmed the new test fails on the base commit and passes with the fix, including under the address and undefined-behaviour sanitizers.

The main request is in the inline comments: put the limit on the whole message, not on each record or option. The parser and the TCP writer both cap a message at 65535 bytes, so this change still produces output the library itself rejects.

• Low — src/lib/record/ares_dns_write.c, in ares_dns_write_buf_tcp(): the goto done after ares_dns_write_buf() is tagged /* LCOV_EXCL_LINE: OutOfMemory */. An oversized message sent over TCP can now reach it through ARES_EBADQUERY, so the tag hides a path that can actually run. Drop the tag.
• Low — docs/ares_dns_record.3, RETURN VALUES: the man page lists only ARES_SUCCESS, ARES_ENOMEM and ARES_EFORMERR for ares_dns_write(). Please add ARES_EBADQUERY, described as "the message is too large to encode".
• Nit — PRs are squash-merged, so the title and body become the commit. The title is too long for one line, which is why GitHub split it. The commit body is only that leftover fragment, with no explanation and no Signed-off-by: line (see CONTRIBUTING.md). Suggested title: ares_dns_write: reject messages larger than 64k.
• Overlap: #1257, an open PR from another author, adds the same per-record check in ares_dns_write_rr() and has had the same request to move it to the whole message. One message-level check covers both PRs.

Thanks, the message-level check is exactly what was needed. I confirmed both new tests fail on the base commit and pass here, including under the address and undefined-behaviour sanitizers, and that the 65535-byte edge case round-trips through ares_dns_parse().

What's left is small: two items from the last round's summary weren't picked up (the coverage-exclusion tag and the man page), and clang-format still needs a pass over the tests.

• Low — src/lib/record/ares_dns_write.c, ares_dns_write_buf_tcp(): the goto done after ares_dns_write_buf() is still tagged /* LCOV_EXCL_LINE: OutOfMemory */. Now that ares_dns_write_buf() returns ARES_EBADQUERY for any message over 65535 bytes, an oversized query sent over TCP reaches this line on a normal path. The tag hides a path that can actually run. Please drop it.
• Low — docs/ares_dns_record.3, RETURN VALUES: the man page still lists only ARES_SUCCESS, ARES_ENOMEM and ARES_EFORMERR for ares_dns_write(). It can now also return ARES_EBADQUERY. Please add it, described as "the message is too large to encode".
• Nit — The reply on the formatting thread says the tests were reformatted, but git clang-format --diff still rewrites both new tests: the declaration alignment in DNSWriteOversizedRdlength and most of the wrapped EXPECT_EQ(...) calls. Please run git clang-format origin/main and commit the result.
• Nit — PRs are squash-merged, so the title becomes the commit subject. At 71 characters it's still long for one line; ares_dns_write: reject messages larger than 64k says the same thing. The PR description now explains the change well.

The two test nits from the last round are fixed: the framing comment now matches the code, and the oversized test reuses one buffer. The writer code hasn't changed since the last round, and both tests still pass here, including under the address and undefined-behaviour sanitizers.

Three items from the previous two rounds are still open. They're all small, but this is the third time they've come up, so please check each one before the next push.

• Low — src/lib/record/ares_dns_write.c, ares_dns_write_buf_tcp(), line 1497: the goto done after ares_dns_write_buf() is still tagged /* LCOV_EXCL_LINE: OutOfMemory */. That tag tells the coverage tool to ignore the line because only an allocation failure can reach it. That is no longer true: ares_dns_write_buf() now returns ARES_EBADQUERY for any message over 65535 bytes, so an oversized query sent over TCP reaches this line on a normal path. Please drop the tag:

  status = ares_dns_write_buf(dnsrec, buf);
  if (status != ARES_SUCCESS) {
    goto done;
  }

• Low — docs/ares_dns_record.3, RETURN VALUES (around line 451): the man page still lists only ARES_SUCCESS, ARES_ENOMEM and ARES_EFORMERR. ares_dns_write() can now also return ARES_EBADQUERY. A public function's man page has to change in the same PR as its behaviour. After the ARES_EFORMERR sentence, add for example:

  \fIares_dns_write(3)\fP also returns
  .B ARES_EBADQUERY
  if the message is too large to encode.
  

• Nit — Formatting: git clang-format --diff origin/main still rewrites both new tests. It changes the declaration alignment at the top of DNSWriteOversizedRdlength (the = nullptr columns are out of line) and nearly every wrapped EXPECT_EQ(...) call in both tests. Running git clang-format origin/main and committing the result fixes all of it at once. Please run the tool itself rather than editing by hand: the last two replies said it was done, but the output hasn't changed.
• Nit — PR title: PRs are squash-merged, so the title becomes the commit subject. At 71 characters it's still long for one line. Suggested: ares_dns_write: reject messages larger than 64k.

All four items from the last round are in. I re-checked them at d2e869e: the coverage-exclusion tag is gone from the goto done in ares_dns_write_buf_tcp(), the ares_dns_record(3) man page lists ARES_EBADQUERY for ares_dns_write(), git clang-format --diff now leaves both tests unchanged, and the title is the short form. Thanks for explaining the clang-format version mismatch.

It builds with warnings as errors and both new tests pass, including under the address and undefined-behaviour sanitizers. The branch merges cleanly with current main. Nothing further from me.