[quinnj]

WebSockets.listen! had no timeouts and no way to set one. A client that connected and never finished its handshake, or went silent after it, held a task and a socket for as long as it liked.

WebSockets.listen!, and WebSockets.serve! with it, now take read_header_timeout, read_timeout and idle_timeout, in seconds or as _ns values, with the meaning and off-by-default values they have in HTTP.serve!:

• read_header_timeout sets a read deadline from accepting the connection to the end of the upgrade request. That covers the TLS handshake on a wss server.
• idle_timeout becomes the session's read idle timeout. The read loop already supports that timeout, re-arming it before each read and closing with 1006 websocket read idle timeout when it expires.
• Both fall back to read_timeout when they are 0, as in HTTP.serve!.

Passing a value in both seconds and _ns form throws ArgumentError, via the same helper serve! uses. The defaults stay 0 (off). Whether to turn them on is a separate decision.

Verification. New testset in test/http_websocket_server_tests.jl that uses already-expired 1 ns timeouts. With read_header_timeout_ns = 1, a client that sends nothing is disconnected. With idle_timeout_ns = 1, the handler's receive throws WebSocketError with reason websocket read idle timeout. On master the keywords do not exist. http_websocket_server_tests.jl (68) and http_websocket_integration_tests.jl pass locally.

Review. The handshake deadline is cleared right after the request is read, so a slow handler or the session is not cut by it. The idle timeout goes through the read_idle_timeout_ns field that the client's read_idle_timeout already uses, so both sides share one code path. HTTP.listen! with WebSockets.upgrade is unchanged: the HTTP server's own deadlines apply there until upgrade clears them. This PR and #1482 both edit _serve_ws_conn! around read_request, so whichever merges second needs a small rebase.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.68%. Comparing base (564fa63) to head (dae821f).
⚠️ Report is 21 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1491      +/-   ##
==========================================
+ Coverage   91.12%   91.68%   +0.56%     
==========================================
  Files          31       31              
  Lines       12535    13208     +673     
==========================================
+ Hits        11422    12110     +688     
+ Misses       1113     1098      -15     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.