[Kayvan-Zahiri]

Problem

The MySQL parameter encoder combines normalized timedelta.days (negative) with the nonnegative seconds and microseconds fields. This changes negative durations when formatting SQL parameters:

from datetime import timedelta
from singlestoredb.mysql import connect
cursor = connect(defer_connect=True).cursor()
cursor.mogrify('SELECT %s', (timedelta(seconds=-1),))
# Before: "SELECT '-1:59:59'"
# After:  "SELECT '-00:00:01'"

A negative microsecond similarly becomes '-1:59:59.999999' instead of '-00:00:00.000001'.

Change

Format the absolute duration and attach its sign once, preserving the existing positive/zero formatting. Add offline public cursor mogrify() regressions and converter round-trip assertions for negative subsecond, second, minute, hour, multi-day, fractional values, and positive/zero controls.

Validation

• Before: 6 negative cases fail; 4 existing-behavior controls pass.
• After: all 10 cases pass, also in SINGLESTOREDB_PURE_PYTHON=1 mode.
• Installed the required pre-commit hook and ran every applicable configured hook against the two changed files; all pass, including Flake8 and MyPy.
• git diff --check passes.

Python 3.12, editable SDK. These tests use deferred connections and never connect to a server. The full MySQL/HTTP protocol suites, accelerated binding path, and management/cloud tests were not run. An exploratory doctest run of unchanged singlestoredb.converters had two existing examples expecting printed None (Python prints nothing for that return value); its other 11 examples passed. No unrelated docstring changes included.

---

Note

Low Risk
Scoped change to timedelta parameter formatting with regression tests; fixes incorrect literals rather than altering unrelated binding behavior.

Overview
Fixes incorrect TIME literals when binding negative datetime.timedelta values. The encoder previously mixed negative days with nonnegative seconds/microseconds, producing wrong strings (e.g. -1:59:59 instead of -00:00:01).

MySQL (escape_timedelta) and HTTP (encode_timedelta) now take the absolute duration, format hours/minutes/seconds as before, and prepend a single - when days < 0. Positive and zero durations are unchanged.

Adds offline tests that assert cursor.mogrify() and encode_timedelta() match expected literals and round-trip through timedelta_fromisoformat.

^{Reviewed by Cursor Bugbot for commit e32e0ef. Bugbot is set up for automated code reviews on this repo. Configure here.}

[Copilot]

🟢 Approval recommended

The localized fix correctly handles negative durations, with focused regression coverage and no unresolved findings.

No open issues 🎉

What changed in this PR

Corrects MySQL parameter encoding for negative timedelta values by formatting the absolute duration and applying the sign once.

Changes:

• Preserves negative duration magnitudes without changing positive or zero formatting.
• Adds offline cursor mogrify() regressions and converter round-trip assertions for signed durations.

File	Description
singlestoredb/​tests/​test_timedelta_parameters.py	Adds negative-duration regressions and positive/zero controls.
singlestoredb/​mysql/​converters.py	Separates the duration’s sign from its magnitude during encoding.

_{Review effort: Balanced}

[kesmit13]

Nice catch. This looks like we inherited this issue from the original PyMySQL code. There is another copy of that converter in singlestoredb/http/connection.py that would need the fix as well.

[Kayvan-Zahiri]

Thanks, good catch. encode_timedelta in singlestoredb/http/connection.py had the same problem, so it gets the same fix in c42b1fc. The test now runs the same cases through the HTTP encoder too; those fail without the change and pass with it.

[Copilot]

🟢 Approval recommended

Both encoder changes correctly preserve sign and magnitude, with focused regression coverage and no unresolved blocking issues.

No open issues 🎉

_{Review effort: Balanced}

[kesmit13]

I think it would be better to name the new test file test_parameters.py. That way if we need to add tests for other parameter types, the filename won't be so specific. Other than that, I can approve. It's still marked as a draft though. Not sure if there is a reason for that.

[Copilot]

🟢 Approval recommended

No active review issues were identified for the signed-duration encoding change.

No open issues 🎉

_{Review effort: Balanced}

[Kayvan-Zahiri]

Renamed to test_parameters.py (329b748). No reason for the draft; it just wasn't flipped.