[LP Bispo (phelipelui)]

Work Item / Issue Reference

GitHub Issue: #863

---

Summary

Cursor.nextset() calls SQLMoreResults and refreshes description and messages, but it leaves rowcount at the value execute() set for the first statement. In a multi-statement batch, the affected-row count of every later INSERT/UPDATE/DELETE was lost. This hits hardest with T-SQL control flow: a WHILE loop sends one result per INSERT, so a tool that sums rowcount over nextset() gets a stale number.

Change (mssql_python/cursor.py, nextset()). After the describe step, rowcount is set the same way execute() sets it:

• -1 when the new result has columns; fetch*() then updates it as it already does;
• DDBCSQLRowCount(self.hstmt) otherwise.

The SQL_NO_DATA path (nextset() returning False) is unchanged. This matches pyodbc, which calls SQLRowCount in Cursor_nextset after SQLMoreResults.

Tests (tests/test_004_cursor.py):

• test_nextset_updates_rowcount runs a batch of INSERT (1 row), INSERT (2 rows), SELECT, UPDATE (3 rows), and DELETE (0 rows). It checks rowcount after each nextset(): 1, 2, -1, 3, 0.
• test_nextset_rowcount_in_loop_batch runs a WHILE loop whose iterations insert 1, 2 and 3 rows. It expects per-result counts [1, 2, 3]. The counts differ on purpose, so a stale rowcount cannot reach the right total by accident.

CHANGELOG: entry under Unreleased → Fixed.

Verification. I ran against SQL Server 2022 (16.0.4305.5) on Python 3.12, Debian 12, with the native bindings from the published 1.16.0 wheel, whose cursor.py is identical to main before this change.

tests/test_004_cursor.py	Without the fix	With the fix
New tests	2 failed (assert 1 == 2, [1, 1, 1] == [1, 2, 3])	2 passed
Whole file	2 failed, 544 passed, 4 skipped	546 passed, 4 skipped

The only failures without the fix are the two new tests, and no other test changes result. Black 26.5.1, the version pinned in .pre-commit-config.yaml, reports both files unchanged.

Out of scope: AsyncCursor.nextset() goes through py-core and was not changed or checked.

🤖 Generated with Claude Code

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[github-actions]

PR Performance Report

Performance could not be assessed.

No matching performance run appeared within the 220-minute wait for 7f5f999e55ccaac099eef7780cdd15ef9cb190ff. No result is available.

[Copilot]

🔵 Needs a closer look

Reported integration results use published-wheel bindings; SQL Server validation against the current native build still needs human confirmation.

No open issues 🎉

What changed in this PR

Addresses #863 by refreshing the synchronous cursor’s rowcount when nextset() advances, aligning it with execute() behavior.

Changes:

• Reports affected-row counts for non-query results and resets rowcount to -1 for result sets.
• Adds regression tests for mixed-statement batches and loops with differing insert counts.
• Documents the fix in the changelog; asynchronous cursors remain unchanged.

File	Description
tests/​test_004_cursor.py	Tests per-result row counts in mixed batches and loops.
mssql_python/​cursor.py	Refreshes rowcount after advancing to a new result.
CHANGELOG.md	Records the fix for #863.

_{Review effort: Balanced}

[LP Bispo (phelipelui)]

@microsoft-github-policy-service agree