[PoisonGodv]

Title: fix(list): bound LTRIM work by list length

Problem

For a string of indices where start == stop and start is beyond the list end, List::Trim does not take the existing empty-range fast path. Its left-delete loop instead runs from metadata.head to metadata.head + start, without being bounded by metadata.tail. On a one-element list this creates point deletes for nonexistent list subkeys.

Minimal example:

RPUSH k a
LTRIM k 2147483647 2147483647

The expected result is a prompt OK with k removed. On the unpatched base, the regression client exceeded a one-second deadline.

A temporary diagnostic counter measured the behavior without allowing an uncontrolled write batch:

• LTRIM k 1000000 1000000 on a one-element list completed after exactly 1,000,000 left-side Delete calls (0.48 seconds in the local ASAN run).
• The maximum-index example reached exactly 1,000,000 left-side Delete calls in 0.27 seconds before the diagnostic deliberately stopped it. The maximum-input run was not allowed to continue, so this is not a claim that billions of completions were observed.

Valkey source aa7545348b688b469d073a01ce4836ed637577f9 returns OK immediately and deletes the key. Its implementation explicitly treats start >= length as an empty range.

Fix

After negative-index normalization, use the existing metadata-deletion path when start >= metadata.size. This is the same result as trimming to an empty list, but avoids constructing deletes for indices that cannot exist. The existing start > stop behavior and retained-range loops are unchanged.

Tests

Add TestLTRIM/LTRIM_with_maximum_start_index, using math.MaxInt32 for both indices and a one-second client deadline. The regression verifies that the resulting list length is zero, the key no longer exists, its TTL is removed (TTL == -2), and the key can be recreated with the expected contents.

Verification on macOS arm64 with Xcode Clang 21 and ASAN:

• New regression on unpatched 84e883225ec2a2de4ed59f86a0f431377face2f2: FAIL (context deadline exceeded). An external watchdog only terminated the test server during cleanup after the client deadline.
• Strengthened regression on the patch: PASS in both txn-context-enabled configurations.
• Complete tests/gocase/unit/type/list package: PASS (TEST_EXIT=0).
• go vet ./unit/type/list: PASS.
• ./x.py check format: PASS.

Not run

• Full C++ unit tests and the complete all-package Go suite.
• clang-tidy and golangci-lint.
• Tests on platforms other than macOS arm64.
• Replication and cluster integration tests.

[github-actions]

Hi @PoisonGodv,

Thank you for your pull request. Please review our Contributing Guide.

Please make sure you understand your changes and explain your reasoning in this pull request. Low-quality pull requests may be closed.

[PoisonGodv]

Could a maintainer please approve the two pending workflows for the current head 7dd48c2? CI (https://github.com/apache/kvrocks/actions/runs/43001672485) and Compatibility Tests (https://github.com/apache/kvrocks/actions/runs/43001672745) are both action_required, so the test suites have not run yet. Thank you!

[PoisonGodv]

@aleksraiden, could you please approve the two pending workflows for the refreshed head b9690334e65de77de9780ec3d98451b5a9e1c1ff? CI (https://github.com/apache/kvrocks/actions/runs/43204693846) and Compatibility Tests (https://github.com/apache/kvrocks/actions/runs/43204693775) are both action_required with zero jobs started, so the test suites have not run yet. Thank you!