ipc (Windows): a server shutdown discards bytes the client has not read (hostd raw terminals may lose the last output)

Aperta
#529 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
69/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
rust

Direzione di ricerca

Start in the Windows implementation of atlas_core::ipc's Stream::Server::poll_shutdown and inspect how it calls NamedPipeServer::disconnect(). Check the Win32 pipe shutdown behavior and add a bounded wait for the client to read the pending bytes or close. Add a cfg(windows) IPC test where a slow client reads all N bytes before EOF; verify it passes, including the existing broken-pipe EOF handling.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

area: runtime type: bug

What

On Windows, atlas_core::ipc's server end discards data the client has not read yet when it shuts down. Stream::Server::poll_shutdown calls NamedPipeServer::disconnect() (DisconnectNamedPipe), and per the Win32 docs that discards anything still in the pipe buffer unless the server first waits for the client (FlushFileBuffers) or for it to close.

Found while fixing the Windows askpass in #525, where it made the password never arrive (3 tests failed on windows-latest: the helper read nothing).

Where else it can bite

crates/atlas-host/src/server.rs, raw terminal connections (attach_raw loop, around L1470-1483 on main at time of writing): when the PTY ends (rrx closes) the loop breaks and calls w.shutdown().await right after the last write_all. On Windows the last chunk of terminal output before the process exits can be lost for a client that has not read it yet. Not reproduced; found by reading the code.

To check: crates/atlas-desktop/src/ctl.rs:123 and debug.rs:145 write a reply and drop the stream (no explicit shutdown). Whether CloseHandle alone loses unread data was not verified.

Proposed fix

Make the semantics safe in one place instead of at each call site: on Windows, poll_shutdown of a server stream waits until the client has read what was written (or has closed) before disconnecting, with a bound. Test under cfg(windows) in ipc tests: server writes N bytes then shuts down; a slow client still reads all N bytes, then EOF. #525 already maps ERROR_BROKEN_PIPE/ERROR_PIPE_NOT_CONNECTED to EOF on the sync client.

Lingua principale
Rust
Stelle
3
Fork
1
Merge medio
9h 51m
PR unite (30g)
310

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di codeatlasdev/atlas

Tutte le issue di codeatlasdev/atlas

Issue simili

Altre issue su Rust

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.