[v0.1.x/Proto] Give stream opening one phase-correct helper instead of three hand-rolled sequences #199

New issue

Closed

opened 2026-09-07 21:21:37 +09:00 by natsukium · 0 comments

natsukium commented

2026-09-07 21:21:37 +09:00

Owner

What

Give the three hand-rolled "open a stream" call sites one shared, phase-correct helper so a caller cannot send a StreamOpener arm without its envelope. felis bridge already has a private open_stream(&msg) (crates/felis-cli/src/cli_bridge.rs); cli_notifications.rs (sessions notifications subscribe) and cli_sessions.rs (rows / search streams) each re-derive the same sequence (allocate the next unopened stream id, send_correlated, register the stream with the driver) with the ordering rule repeated inline.

Why it is not part of #196

#196 makes CheckedFrame::encode refuse a StreamOpener sent without an envelope, so the mistake becomes a typed local error instead of a silent EOF. It does not give callers the right way to do it. That helper cannot be a single generic Connection::open_stream: Driver::observing() is Observer-side and Phase::Setup-only (it debug-asserts on the phase), while Region::Rows and Search::Query open streams from Attached. The shape is therefore per surface (or a subscribe_notifications verb on Connection beside the existing attached-phase openers), and hoisting the bridge helper touches three CLI call sites with a control-surfaces/cli doc cascade of its own.

Acceptance criteria

One helper (or one per phase) in felis-client-core owns "next unopened stream id + send_correlated + register"; cli_bridge.rs, cli_notifications.rs, and cli_sessions.rs call it and the inline ordering comments disappear.
A test proves the helper allocates ids in the driver's expected order from both Setup (Observer) and Attached, and that the raw sequence is no longer reachable from the CLI crate.
Docs cascade: docs/reference/ipc.md names the helper where it describes stream opening from the client side; no wire, CLI, or config change, so no CHANGELOG.

Split from #196.

## What Give the three hand-rolled "open a stream" call sites one shared, phase-correct helper so a caller cannot send a `StreamOpener` arm without its envelope. `felis bridge` already has a private `open_stream(&msg)` (`crates/felis-cli/src/cli_bridge.rs`); `cli_notifications.rs` (`sessions notifications subscribe`) and `cli_sessions.rs` (rows / search streams) each re-derive the same sequence (allocate the next unopened stream id, `send_correlated`, register the stream with the driver) with the ordering rule repeated inline. ## Why it is not part of #196 #196 makes `CheckedFrame::encode` refuse a `StreamOpener` sent without an envelope, so the mistake becomes a typed local error instead of a silent EOF. It does not give callers the right way to do it. That helper cannot be a single generic `Connection::open_stream`: `Driver::observing()` is Observer-side and `Phase::Setup`-only (it debug-asserts on the phase), while `Region::Rows` and `Search::Query` open streams from `Attached`. The shape is therefore per surface (or a `subscribe_notifications` verb on `Connection` beside the existing attached-phase openers), and hoisting the bridge helper touches three CLI call sites with a control-surfaces/cli doc cascade of its own. ## Acceptance criteria - [ ] One helper (or one per phase) in `felis-client-core` owns "next unopened stream id + `send_correlated` + register"; `cli_bridge.rs`, `cli_notifications.rs`, and `cli_sessions.rs` call it and the inline ordering comments disappear. - [ ] A test proves the helper allocates ids in the driver's expected order from both `Setup` (Observer) and `Attached`, and that the raw sequence is no longer reachable from the CLI crate. - [ ] Docs cascade: `docs/reference/ipc.md` names the helper where it describes stream opening from the client side; no wire, CLI, or config change, so no CHANGELOG. Split from #196.

natsukium referenced this issue

2026-09-07 21:21:38 +09:00

protocol: FrameWriter::send accepts a correlated arm and writes a frame the peer must refuse #196

natsukium added the

priority/P2

label

2026-09-07 21:21:39 +09:00

natsukium referenced this issue from a pull request that will close it,

2026-09-08 22:18:00 +09:00

client-core: one phase-correct helper for opening a stream #235

natsukium closed this issue

2026-09-08 22:41:24 +09:00