WebSocket close codes inconsistent across Dart versions and affected by pingInterval setting

[coderabbitai]

Summary

WebSocket close codes are inconsistent across different Dart versions and unexpectedly affected by the pingInterval setting when the connection is closed abnormally (e.g., by killing the server process).

Expected Behavior

When a WebSocket connection is closed abnormally (server process killed), the close code should consistently be 1006 (Abnormal) across all Dart versions, regardless of whether pingInterval is set.

Actual Behavior

With pingInterval set client-side:

• Dart 3.5.0, 3.6.2, 3.8.1: Returns 1001 (Going away)

Without pingInterval set client-side:

• Dart 3.7.2: Returns 1005 (No status)
• Dart 3.8.1: Returns 1006 (Abnormal)

Issues

1. Version inconsistency: Different Dart versions return different close codes for the same scenario
2. Unexpected pingInterval effect: Setting pingInterval client-side changes the close code behavior, which seems unrelated to the close code logic

Context

This issue was discovered while writing tests for WebSocket behavior where the server process is terminated abruptly. The inconsistent behavior makes it difficult to write reliable tests that work across different Dart versions.

Backlinks

• Original PR: refactor!: Replace DuplexStreamChannel with RelicWebSocket serverpod/relic#91
• Comment discussion: refactor!: Replace DuplexStreamChannel with RelicWebSocket serverpod/relic#91 (comment)

Reported by: @nielsenko

[mraleph]

Please provide complete runnable reproduction. Also please don't use AI bots to file issues - because my first reaction would be just to close and ignore them.

[nielsenko]

@mraleph Sorry, this should just have been a reminder issue in our own repo. Still trying to trying to get a hang of all these new ai tools. I'm afraid I cannot close it myself. Could I ask you to close it. I'm very sorry for the inconvinience.

[mraleph]

FWIW the difference between Dart 3.7.2 and Dart 3.8.1 is intentional and originates from 54c8719

I think that the difference with pingInterval might also be intentional: there certainly can be a race between detecting connection issues via ping-pong vs detecting connection issues at a lower level of the stack (e.g. you kill the server which forces OS to shutdown connection and send a FIN packet).