test: prune stale entries from expectations.txt

scripts/runner.node.mjs

@@ -372,6 +372,7 @@ const skipsForLeaksan = (() => {
   }
   return readFileSync(path, "utf-8")
     .split("\n")
+    .map(line => line.trim())
     .filter(line => !line.startsWith("#") && line.length > 0);
 })();
 

test/expectations.txt

@@ -17,91 +17,21 @@
 # Entries match by SUBSTRING, so a path prefix quarantines an entire family.
 # A stale entry is invisible: the file simply never runs and nobody finds out.
 
-# Tests that are broken
-[ WINDOWS-AARCH64 ] test/js/node/test/parallel/test-repl-close.js [ FAIL ] # EPIPE on stdin.write to closed child process
-
 # The whole file's beforeAll builds a napi addon with node-gyp, and the Windows
 # agents have no ClangCL toolset ("MSB8020: The build tools for ClangCL cannot
 # be found"), so nothing in the file can run there. Remove once the agent image
-# ships ClangCL. Runs everywhere else.
+# ships ClangCL. Runs everywhere else. (Verified still failing, build 87834.)
 [ WINDOWS ] test/bundler/native-plugin.test.ts [ SKIP ] # node-gyp needs a ClangCL toolset the Windows agents do not have
 
-# Verbatim node v26.3.0 test asserting a FinalizationRegistry callback fires
-# within ONE globalThis.gc() + ONE setImmediate after the connect callback's
-# closure is unreferenced. The FinalizationRegistry spec gives no timing
-# guarantee for cleanup callbacks; JSC schedules them via DeferredWorkTimer
-# with no defined ordering relative to the immediate queue. The connect
-# listener IS removed (verified: listenerCount("secureConnect") === 0 in
-# done()) and the object IS collected (test passes 70/70 on darwin and
-# glibc Linux); on alpine x64 the FR callback delivery slips past the single
-# setImmediate after this PR's added module loads at process startup shift
-# the heap layout. The robust fix is gcUntil() rather than a single tick,
-# but the file is a verbatim upstream port. Quarantined on the failing
-# linux-x64-musl matrix only; still runs everywhere else (build 63145:
-# alpine 3.23 x64 + x64-baseline only).
-[ LINUX-X64-MUSL ] test/js/node/test/parallel/test-tls-connect-memleak.js [ FLAKY ] # JSC FinalizationRegistry callback delivery vs setImmediate timing on musl x64
-
-# Both tests mock _handle.setKeepAlive and assert it receives SECONDS
-# (libuv's uv_tcp_keepalive convention). In Bun, _handle is the public
-# Bun.Socket whose setKeepAlive is documented in MILLISECONDS, so net.ts
-# forwards ms. End-to-end kernel coverage (TCP_KEEPIDLE=4 for 4000ms on
-# both Bun.connect and node:net) lives in test/js/bun/net/socket.test.ts.
-test/js/node/test/parallel/test-net-connect-keepalive.js [ FAIL ] # asserts _handle.setKeepAlive gets seconds; Bun's _handle is Bun.Socket (ms)
-test/js/node/test/parallel/test-net-server-keepalive.js [ FAIL ] # asserts _handle.setKeepAlive gets seconds; Bun's _handle is Bun.Socket (ms)
-
-# Vendored node v26.3.0 stream tests blocked on missing native subsystems (see PR #31826)
-test/js/node/test/parallel/test-stream-wrap.js [ FAIL ] # needs internal/test/binding + js_stream (net.Socket({handle}) libuv compat layer)
-test/js/node/test/parallel/test-stream-wrap-drain.js [ FAIL ] # needs internal/js_stream_socket (net.Socket({handle}) libuv compat layer)
-test/js/node/test/parallel/test-stream-wrap-encoding.js [ FAIL ] # needs internal/js_stream_socket (net.Socket({handle}) libuv compat layer)
-
-# Spawns test-http-max-http-headers.js, which is not vendored, and asserts its
-# exit code. Only "passed" before #32631 because the old node:test shim never
-# awaited done-style callbacks, so the cp.on('close') assertions never ran.
-# With the real (t, done) support, the second/fourth tests fail (child exits 1
-# with "Module not found" where exit 0 is expected). Remove this entry once
-# test-http-max-http-headers.js is vendored.
-test/js/node/test/parallel/test-set-http-max-http-headers.js [ FAIL ] # spawns test-http-max-http-headers.js which is not vendored
-
-# Tests failed due to ASAN: attempting free on address which was not malloc()-ed
-[ ASAN ] test/integration/next-pages/test/dev-server-ssr-100.test.ts [ CRASH ]
-[ ASAN ] test/integration/next-pages/test/next-build.test.ts [ CRASH ]
-[ ASAN ] test/js/third_party/next-auth/next-auth.test.ts [ CRASH ]
-
-# Tests failed due to ASAN: SEGV on unknown address
-[ ASAN ] test/integration/next-pages/test/dev-server.test.ts [ CRASH ]
-
-# Tests failed due to ASAN: use-after-poison
-[ ASAN ] test/napi/napi.test.ts [ CRASH ] # can throw an exception from an async_complete_callback
-
-# Tests failed due to ASAN: unknown-crash
-[ ASAN ] test/js/sql/tls-sql.test.ts [ CRASH ] # After: Throws on illegal transactions
-
-# Tests timed out due to ASAN
-[ ASAN ] test/js/bun/spawn/spawn.test.ts [ TIMEOUT ]
-[ ASAN ] test/cli/inspect/inspect.test.ts [ TIMEOUT ]
-
-# Tests failed due to memory leaks
-[ ASAN ] test/js/node/url/pathToFileURL.test.ts [ LEAK ] # pathToFileURL doesn't leak memory
-[ ASAN ] test/js/node/fs/abort-signal-leak-read-write-file.test.ts [ LEAK ] # should not leak memory with already aborted signals
-[ ASAN ] test/js/web/streams/streams-leak.test.ts [ LEAK ] # Absolute memory usage remains relatively constant when reading and writing to a pipe
-[ ASAN ] test/cli/run/require-cache.test.ts [ LEAK ] # files transpiled and loaded don't leak file paths > via require()
-
 # Windows-only gap in named-pipe half-close: WindowsNamedPipe::shutdown() /
 # on_read_error(EOF) issue a full uv_close instead of uv_shutdown, so the
 # final write after the peer half-closes is lost. Passes on Linux and macOS.
-[ WINDOWS ] test/js/node/test/parallel/test-net-pingpong.js [ FAIL ] # named-pipe half-close (FIN) handling
-[ WINDOWS ] test/js/node/test/parallel/test-net-server-listen-path.js [ FAIL ] # EADDRINUSE not reported for a second listen on the same pipe path
-
-# Same Windows half-close + client-RST gap over a node:http server: the server
-# half-closes (res.socket.end()) mid-upload, the fetch client RSTs the cut-short
-# 10 MB POST body, and that RST is not surfaced to the server's still-open
-# readable side on Windows, so the accepted connection never ends and
-# server.close() (via `await using`) waits forever. The assertion itself passes;
-# only the teardown hangs. Passes on Linux and macOS.
+# (Verified still failing on both Windows lanes, build 87834.)
+[ WINDOWS ] test/js/node/test/parallel/test-net-pingpong.js [ FAIL ] # named-pipe half-close: count 1000 !== 1001 at close
 
 # Cluster workers sharing a listen port behave differently on Linux, where
 # SO_REUSEPORT load-balances across the workers' own listeners instead of the
 # primary distributing accepted connections; the upstream test's expectations
-# only hold on the distributing model. Passes on macOS, Windows and FreeBSD.
+# only hold on the distributing model. Passes on macOS and Windows.
+# (Verified still failing on every Linux lane, build 87834.)
 [ LINUX ] test/js/node/test/sequential/test-net-listen-shared-ports.js [ FAIL ] # SO_REUSEPORT shared-listener semantics on Linux
-[ DARWIN ] test/js/bun/io/fetch/fetch-abort-slow-connect.test.ts [ FLAKY ] # connect-during-abort race depends on host routing of 192.0.2.1; darwin CI agents return an immediate routing error