Release 3.0.0-preview.6

[mtmk]

Preview release on the 3.0 line bumping version to 3.0.0-preview.6. Syncs everything merged to main since preview.5: a new server error event on NatsConnection, slnx solution-format migration, README and example docs additions, and a CI matrix bump to nats-server v2.12. No 3.0-only changes.

From main:

• Add server error event (Add server error event #745)
• Migrate solution to slnx format (Migrate solution to slnx format #1131)
• Add Client and Orbit section to README (Add Client and Orbit section to README #1133)
• Add Example Docs (Add Example Docs #1119)
• ci: test against v2.12, drop v2.10 (Add v2.12 to test matrix, drop v2.10 #1136)

[synadia-claude-reviewer]

Claude finished @mtmk's task in 2m 29s —— View job

---

PR Review: Release 3.0.0-preview.6

• Gather context and understand request
• Read PR diff and changed files
• Review server error event implementation
• Review async/await correctness and thread safety
• Review API surface and nullable annotations
• Post findings

---

Summary

The main functional addition is the ServerError event on NatsConnection / INatsConnection. The rest of the diff is docs, examples, CI updates, and the slnx migration. No 3.0-specific changes.

The implementation is solid overall — it follows the existing async event-channel pattern correctly, error strings are never executed, and thread safety is consistent with the rest of the event dispatch path.

A couple of observations worth noting:

---

NatsServerErrorEventArgs.Error nullability contract

Error is typed string (non-nullable), but the unit test bypasses the type system with null! and asserts Assert.Equal(error, args.Error) — which actually stores null in a non-nullable property.

From the production call path (ParseError) this can never happen — ParseError always returns a string (either string.Empty or a decoded UTF-8 slice). So the null test is a false-positive edge case that the constructor's public API will never receive in real use. Two reasonable fixes:

1. Make Error nullable (string?) and update the test to be Assert.Equal(error ?? string.Empty, args.Error) if you want to guard against misuse.
2. Remove the null test case since the constructor is only ever called from the protocol processor with a non-null result.

Option 2 is simpler given the production reality.

Fix this →

---

Event ordering: PushEvent before TrySetObservedException

In NatsReadProtocolProcessor.cs lines 407–408 and 415–416, the event is pushed to the channel before the pong-signal exception is set:

_connection.PushEvent(NatsEvent.ServerError, new NatsServerErrorEventArgs(error));
_waitForPongOrErrorSignal.TrySetObservedException(new NatsServerException(error));

Because PushEvent only enqueues (non-blocking) and the actual ServerError handler fires asynchronously from the event loop, the TrySetObservedException will almost certainly resolve first — reconnect logic may begin before any registered ServerError handler runs. This is the same behavior as all other events in the system, so it's acceptable, but worth documenting if users ask why the event arrives "after" reconnection has started.

---

ParseKind prefix ordering: no bug

The order "Maximum Account Active Connections" before "Maximum Connections" is safe: the two prefixes are mutually exclusive, so no wrong branch is taken regardless of ordering. ✓

---

CI: test_natsiodocs.yml always uses nats-server@latest

The new workflow installs the server via go install github.com/nats-io/nats-server/v2@latest. This is intentional for docs examples but it does mean any breaking server change in main could silently break the docs CI without a pinned version. This is a deliberate trade-off and fine for examples.

---

Overall the change is correct and ready to merge. The null-in-Error observation above is the only thing worth a follow-up (even if as a deferred cleanup).

[github-actions]

🚀 Deployed on https://6a0454208a0b352c2a8fb92e--teal-klepon-7e92d0.netlify.app

[synadia-claude-reviewer]

null is passed with ! suppression, but NatsServerErrorEventArgs.Error is typed string (non-nullable). Since ParseError in NatsReadProtocolProcessor never returns null (it returns string.Empty for short sequences, a decoded UTF-8 string otherwise), this test case can't happen in production and silently stores null in a non-nullable property. Consider removing the null case here.

[mtmk]

Closing in favour of a clean PR. The diff here picked up all of the main sync because release/3.0 hadn't been pushed with the merge before this branch was cut; reopening with just the version.txt bump on top of the now-synced release/3.0.