go/src/net/http/clientserver_test.go, branch main

net/http: temporarily disable HTTP/3 tests

2026-04-14T01:39:04Z

Tests for HTTP/3 are unfortunately still flaky. The flakes seem to come mostly from two sources: 1. QUIC connections are not shutdown fast enough, causing them to be detected as goroutine leaks. 2. x/net/quic has a race condition where its output buffer are written to without locks (synchronously and intentionally, for optimization), but may be asynchronously written when the QUIC stream receives a STOP_SENDING frame. Therefore, skip http3Mode for now until the above flakes are fixed. For #70914 Fixes #78701 Change-Id: If0c857c0933b977fb10a4481124680726a6a6964 Reviewed-on: https://go-review.googlesource.com/c/go/+/766740 Reviewed-by: Nicholas Husin Auto-Submit: Nicholas Husin LUCI-TryBot-Result: golang-scoped@luci-project-accounts.iam.gserviceaccount.com Reviewed-by: Carlos Amedee

net/http: fix wrong context being used when shutting down HTTP/3 server

2026-04-13T19:51:05Z

t.Context was accidentally used within a t.Cleanup. This is a mistake since t.Context will always have been cancelled by the time t.Cleanup runs. This is likely the reason for flakiness in some builders: our HTTP/3 finishes before QUIC connections can finish closing asynchronously, which are then detected as goroutine leaks. To account for this, also make the shutdown context timeout more generous. For #70914 Change-Id: I0e5f06a47ef25c5df535543fc1e602f16a6a6964 Reviewed-on: https://go-review.googlesource.com/c/go/+/766620 LUCI-TryBot-Result: golang-scoped@luci-project-accounts.iam.gserviceaccount.com Reviewed-by: Nicholas Husin Reviewed-by: Damien Neil

net/http: run tests for HTTP/3 where it can already pass

2026-04-10T21:18:34Z

By default, our test harnesses (run and runSynctest) now use http3Mode, in addition to http1Mode and http2Mode, when no []testMode were explicitly defined for a given test. Tests that cannot currently pass for HTTP/3 have been modified to use http3SkippedMode, which serves as a convenient alias for the old default of []testMode{http1Mode, http2Mode}. We changed the default mode and defined http3SkippedMode so we have a clear list of TODOs in terms of how much changes are still needed before our HTTP/3 implementation reaches basic feature parity with HTTP/1 and HTTP/2. For #70914 Change-Id: I719d5d66399a51f7c3d96180ebed9b606a6a6964 Reviewed-on: https://go-review.googlesource.com/c/go/+/765320 Reviewed-by: Damien Neil LUCI-TryBot-Result: Go LUCI Reviewed-by: Nicholas Husin

net/http: add support for running HTTP tests against HTTP/3

2026-04-10T15:24:28Z

Add support within clientserver_test.go to bring up a test HTTP/3 server and client when http3Mode testMode option is passed. To be able to reuse net/http/httptest, net/http/httptest.Server.StartTLS (and Start) have been modified so they can be called with a nil Listener. In such cases, both methods will behave identically as usual, but will not actually make its server serve or set its transport dialer, both of which requires having a listener. This should be a no-op for regular users of the package, whose entrypoint via functions such as NewServer will automatically set a local listener. Actually enabling HTTP/3 for our tests will be done in a separate CL. For #70914 Change-Id: Ibc5fc83287b6a04b46e668a54924761a92b620a4 Reviewed-on: https://go-review.googlesource.com/c/go/+/740122 Reviewed-by: Damien Neil Reviewed-by: Nicholas Husin LUCI-TryBot-Result: Go LUCI

net/http: use net/http/internal/http2 rather than h2_bundle.go

2026-03-12T15:13:20Z

Rework net/http/internal/http2 to use internally-defined types rather than net/http types (to avoid an import cycle). Remove h2_bundle.go, and replace it with calls into net/http/internal/http2 instead. For #67810 Change-Id: I56a1b28dbd0e302ab15a30f819dd46256a6a6964 Reviewed-on: https://go-review.googlesource.com/c/go/+/751304 Reviewed-by: Nicholas Husin LUCI-TryBot-Result: Go LUCI Auto-Submit: Damien Neil Reviewed-by: Nicholas Husin

net/http: add Transport.NewClientConn

2025-11-25T01:26:36Z

For #75772 Change-Id: Iad7607b40636bab1faf8653455e92e9700309003 Reviewed-on: https://go-review.googlesource.com/c/go/+/722223 Reviewed-by: Nicholas Husin Reviewed-by: Nicholas Husin Auto-Submit: Damien Neil LUCI-TryBot-Result: Go LUCI

net/http: add more tests of transport connection pool

2025-09-26T22:06:09Z

Add a variety of addtional tests exercising client connection pooling, in particular HTTP/2 connection behavior. Change-Id: I7609d36db5865f1b95c903cfadb0c3233e046c09 Reviewed-on: https://go-review.googlesource.com/c/go/+/615896 Reviewed-by: Nicholas Husin Reviewed-by: Nicholas Husin LUCI-TryBot-Result: Go LUCI Auto-Submit: Damien Neil

net/http: use synctest.Test rather than Run

2025-05-21T23:06:11Z

Use the non-experimental Test function. As a bonus, this lets us drop the hacks we were doing to support t.Cleanup inside bubbles. Change-Id: I070624e1384494e9d5fcfee594cfbb7680c1beda Reviewed-on: https://go-review.googlesource.com/c/go/+/675315 LUCI-TryBot-Result: Go LUCI Auto-Submit: Damien Neil Reviewed-by: Jonathan Amsterdam

net/http: use runtime.AddCleanup instead of runtime.SetFinalizer

2025-02-14T17:59:55Z

Replace the usage of runtime.SetFinalizer with runtime.AddCleanup in tests. Updates #70907 Change-Id: Idd3f1c07f6a7709352ca09948fbcb4a0ad9418bb Reviewed-on: https://go-review.googlesource.com/c/go/+/648655 Auto-Submit: Carlos Amedee Reviewed-by: Damien Neil LUCI-TryBot-Result: Go LUCI

net/http: test for racing idle conn closure and new requests

2024-11-26T18:05:09Z

TestTransportRemovesH2ConnsAfterIdle is experiencing flaky failures due to a bug in idle connection handling. Upon inspection, TestTransportRemovesH2ConnsAfterIdle is slow and (I think) not currently testing the condition that it was added to test. Using the new synctest package, this CL: - Adds a test for the failure causing flakes in this test. - Rewrites the existing test to use synctest to avoid sleeps. - Adds a new test that covers the condition the test was intended to examine. The new TestTransportIdleConnRacesRequest exercises the scenario where a never-used connection is closed by the idle-conn timer at the same time as a new request attempts to use it. In this race, the new request should either successfully use the old connection (superseding the idle timer) or should use a new connection; it should not use the closing connection and fail. TestTransportRemovesConnsAfterIdle verifies that a connection is reused before the idle timer expires, and not reused after. TestTransportRemovesConnsAfterBroken verifies that a connection is not reused after it encounters an error. This exercises the bug fixed in CL 196665, which introduced TestTransportRemovesH2ConnsAfterIdle. For #70515 Change-Id: Id23026d2903fb15ef9a831b2df71177ea177b096 Reviewed-on: https://go-review.googlesource.com/c/go/+/631795 Reviewed-by: Jonathan Amsterdam LUCI-TryBot-Result: Go LUCI Auto-Submit: Damien Neil