[dhensby]

What this does:

msnodesqlv8 5.6.0 (2026-10-11) changed how errors, cancels and info messages reach event consumers. The floating msnodesqlv8@^5 CI step turned master red the next day, in run 38225080240, with no change to our code. The failures were all 16 Windows Node 22/24 jobs, and only the msnodesqlv8 v5 step.

setup-sqlserver had no part in it. Its SHA, Native Client and SQL Server builds were the same in the green and the red run, and the tedious and msnodesqlv8@^4 steps passed in the same jobs.

The CI failures pointed at real bugs for anyone installing msnodesqlv8 today.

Each commit fixes one of them:

1. fix: reject msnodesqlv8 requests whose errors do not end the batch
   • In 5.6 an error that does not end the batch arrives with more=true and the query carries on to 'done' (fix: prints, errors and results after a failed statement in a batch were lost TimelordUK/node-sqlserver-v8#475). The adapter only failed on a more=false error, so these requests resolved: execute() of a missing procedure, RAISERROR followed by results, and a PreparedStatement whose prepare fails.
   • Requests now reject in 'done' when errors were collected, as on tedious.
   • This commit adds stub-driven unit tests for the adapter. They stand a fake in for require('msnodesqlv8') and play the events each version emits, so this logic is tested on every CI job and locally, not just on Windows.
2. fix: keep every preceding error on msnodesqlv8 requests: precedingErrors kept only one earlier error. Batches now report several, so all are kept, as on tedious.
3. fix: reject a terminated msnodesqlv8 statement with its own error
   • 5.6 emits 3621 "The statement has been terminated." before the error it follows. A duplicate-key insert therefore rejected with 3621 instead of 2627, which breaks err.number checks.
   • The Handle queries that are terminated before all errors are handled #1145 behaviour for msnodesqlv8 2.0 (failed query() is never settled #1126), which can stop after the 3621, is kept.
4. fix: report cancelled msnodesqlv8 requests, including paused streams, as ECANCEL
   • 5.6 can cancel a running query. It emits 'done' as soon as the cancel is through, then a late HY008 Operation canceled error.
   • The adapter took that 'done' as success. So a cancelled request resolved with partial results, and a stream emitted an EREQUEST after 'done'.
   • A cancel now reports ECANCEL. Streams get an 'error' and then 'done' with what was read, as on tedious; this also covers a paused stream on any msnodesqlv8 version.
   • Anything msnodesqlv8 emits after 'done' is ignored, but an 08S01 still marks the connection.
   • The README documents both points.
5. ci: correct the note on msnodesqlv8 v5 prebuilds: the note claimed ^5 could never load on Node 20 again; since 5.3.0 msnodesqlv8 ships a single Node-API binary. This commit only changes the comment.
   • Node 22/24 deliberately keep floating on msnodesqlv8@^5, so CI meets each new release when users do. That is how 5.6.0 was caught.
   • An exact pin was tried and dropped in review.
   • The old event behaviour stays covered by v4, and by 5.2.1 on Node 20.

Behaviour change to note:

• Cancelling a paused msnodesqlv8 stream used to end with a bare 'done'. On every msnodesqlv8 version it now emits 'error' (ECANCEL) first, which matches tedious and the README.
• pipe() / toReadableStream() consumers without an error listener will now see that error, as they already do on tedious.

Known limits: these can't be fixed in the adapter and have been reported upstream in TimelordUK/node-sqlserver-v8#479.

• On 5.6 a cancel that the driver silently refuses still reports 'done' and success, so it is reported here as ECANCEL.
• A cancel racing a query's completion on msnodesqlv8 below 5.6 can report ECANCEL.

Testing:

• Lint, the unit tests (159, including 21 new stub-driven msnodesqlv8 tests) and the CLI tests pass locally.
• The tedious integration suite passes locally against Azure SQL Edge. The exception is user defined types, which always fails there because SQL Edge has no CLR.
• The new and changed integration tests cover a non-batch-ending error, a constraint violation (2627 not 3621) and a running-query cancel. The transaction test that used to hang for 30s, and then stalled the after all hook, now fails fast.
• The msnodesqlv8 behaviour was checked against the 2.7.0, 3.1.0, 4.5.0, 5.2.1, 5.5.x and 5.6.0 sources.
• Windows CI is the first real run against the driver itself. Two things to watch:
  • The constraint-violation case where the error follows an earlier result is gated to msnodesqlv8 >= 5.1.9; older versions lose that error (ODBC errors not bubbling up when PRINT + THROW are used. TimelordUK/node-sqlserver-v8#396).
  • The running-query cancel test expects ECANCEL on 5.6 and a completed query before it.

Related issues:

Upstream reports for the 5.6.0 behaviour:

• 5.6.0: a failed statement's info messages are emitted before its error (3621 "The statement has been terminated." arrives before 2627) TimelordUK/node-sqlserver-v8#478: info messages emitted before the error they follow (3621 before 2627).
• 5.6.0: on cancel, 'done' is emitted before the query has ended, and also when nothing was cancelled TimelordUK/node-sqlserver-v8#479: on cancel, 'done' is emitted before the query has ended, and also when nothing was cancelled.
• After a query fails, the connection's work queue advances once too often, so later queries can overlap (5.1.0 – 5.6.0) TimelordUK/node-sqlserver-v8#481: after a query fails, msnodesqlv8's per-connection work queue advances once too often. This is the one way found for a connection released here, on a cancel or an error, to overlap the next borrower's query. It predates this PR.

The other change, TimelordUK/node-sqlserver-v8#475 (an error that does not end the batch arrives with more=true, then 'done'), was not reported. It is intended and documented: the v5.6.0 release notes call it "the main thing to warn about". Commit 1 adapts to it.

Pre/Post merge checklist:

• Update change log

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

A cancel-generated done can release a connection while the driver is still using it, allowing unsafe concurrent reuse.

2 open findings

• Stub omits final free event and permits premature release · New
• Fix ungrammatical wording around dropped HY008 events · New

What changed in this PR

Adapts the msnodesqlv8 adapter to version 5.6 event ordering and cancellation behavior.

Changes:

• Preserves and reports batch errors correctly.
• Reports cancellations as ECANCEL.
• Adds event-stub and integration coverage plus documentation updates.

File	Description
lib/​msnodesqlv8/​request.js	Handles revised error, info, and cancellation events.
test/​common/​unit.js	Adds msnodesqlv8 event-stub tests.
test/​common/​tests.js	Adds shared error and cancellation integration tests.
test/​msnodesqlv8/​msnodesqlv8.js	Enables version-aware msnodesqlv8 tests.
test/​tedious/​tedious.js	Runs new shared tests against tedious.
README.md	Documents cancellation behavior.
.github/​workflows/​nodejs.yml	Corrects the prebuild compatibility note.
.github/​copilot-instructions.md	Documents the event-stub testing convention.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟢 Approval recommended

The version-specific event behavior is handled consistently and covered by focused unit and integration tests.

0 open findings

2 resolved since last review

• Stub omits final free event and permits premature release
• Fix ungrammatical wording around dropped HY008 events

🧠 Review effort: Balanced

[Copilot]

🔵 Needs a closer look

Native-driver event ordering, cancellation races, and version compatibility warrant final human validation on Windows CI.

0 open findings

🧠 Review effort: Balanced

[github-actions]

🎉 This PR is included in version 12.7.5 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀