diff options
| author | Igor Soarez <igor@soarez.org> | 2026-11-20 16:38:11 +0200 |
|---|---|---|
| committer | Igor Soarez <igor@soarez.org> | 2026-11-20 16:38:11 +0200 |
| commit | 31eee500cd9c570c960bbb9a50209c0c21c1b716 (patch) | |
| tree | 68c3fd9a7a093493b821552a76f8b55e84a41591 /test-suite/unit/grant-scope.test.mjs | |
| parent | 7658ed71898c7daca2661c7d15ffe03c8608c865 (diff) | |
server.close() refuses new connections but leaves open the ones busy at
that moment, such as the extension posting a result. When the owner pi
reloaded, Chrome sent the extension's next /next on such a connection,
and the stopped bridge answered it, and every poll after it, for good.
Commands went to the new owner, which never heard from Chrome: "not
polling (last seen never)" until the extension was reloaded. Seen
2026-11-20 with lsof: Chrome connected to a pi that no longer listened.
stop() now closes every connection once the final answers are written,
so the extension reconnects to whoever owns the port next.
bridge-handover.test.mjs covers a connection busy at the stop and one
opened before it; both reached the stopped bridge before this change.
Diffstat (limited to 'test-suite/unit/grant-scope.test.mjs')
| -rw-r--r-- | test-suite/unit/grant-scope.test.mjs | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/test-suite/unit/grant-scope.test.mjs b/test-suite/unit/grant-scope.test.mjs index 7717131..fd67508 100644 --- a/test-suite/unit/grant-scope.test.mjs +++ b/test-suite/unit/grant-scope.test.mjs @@ -21,7 +21,7 @@ function piProcess(env) { const fakeProcess = Object.create(process, { env: { value: { ...env, PI_CHROME_BRIDGE_PORT: "0" } } }); const context = vm.createContext({ process: fakeProcess, console, Buffer, URL, AbortController, fetch, - setTimeout, clearTimeout, setInterval, clearInterval, + setTimeout, clearTimeout, setInterval, clearInterval, setImmediate, }); let runtime; // A grant an earlier runtime left on globalThis, e.g. a build that saved the env grant there. |
