mirror of
https://github.com/Hmbown/DeepSeek-TUI.git
synced 2026-09-02 22:36:19 +08:00
`streaming/chunking.rs` carried ~460 lines and eight hysteresis constants that
could not change what the user saw. `DrainPlan::Single` was constructed
nowhere, and all three `decide()` return paths yielded `DrainPlan::Available`,
so `run_commit_tick` always drained the whole queue. The mode the policy
tracked fed exactly one `tracing::trace!` and a test-only accessor. Reduced
motion routed `MotionPolicy::as_low_motion` into it every frame, which
likewise changed nothing.
`LineBuffer` was in the same state. Its docs promised a partial code fence
could never reach the renderer, but both `StreamingState` constructors set
`bypass_gate: true`, so nothing was ever pushed through it. That protection is
real one layer down -- `ParseState::commit_complete_lines`
(tui/markdown_render.rs) commits only whole lines and re-parses the trailing
partial line each tick -- so the gate is deleted rather than switched on:
enforcing it would hold assistant prose until a newline arrives, which is the
chunkiness the 16 ms beat exists to remove.
What remains is what was actually running: `StreamBuffer` accumulates raw
deltas and a commit beat takes everything received since the previous beat.
`StreamChunker`'s per-delta `VecDeque` and enqueue timestamps existed only to
build snapshots for the dead policy, and its `pending` field was never
written. The per-block `MarkdownStreamCollector` survived only as an
`is_thinking` / `is_streaming` flag holder once the gate went, and
`start_text` / `start_thinking` took a `width` that every caller passed as
`None`.
`StreamDisplayClock` is deliberately unchanged, including
`note_delta_with_backlog`: that path feeds the clock's own catch-up, not the
deleted policy. It is still staged rather than live (every drain site passes
queued = 1) and docs/MOTION_CONTRACT.md keeps saying so; the doc also stops
claiming ~30 FPS now that the beat is 16 ms.
The three deleted files were swept into 7d425158c by a concurrent lane while
they sat staged; this commit carries the rest of the change.
59 lines
3.0 KiB
Markdown
59 lines
3.0 KiB
Markdown
# Motion contract
|
|
|
|
Central motion policy for the underwater TUI lives in
|
|
`crates/tui/src/tui/motion/`.
|
|
|
|
## Modes
|
|
|
|
| Mode | Decorative ambient | Status spinner | Streaming |
|
|
|------|--------------------|----------------|-----------|
|
|
| `Full` | yes | animated braille | steady ~60 FPS display clock (16 ms); catch-up is STAGED, not live — see note below |
|
|
| `Reduced` | no | static calm glyph | **same** display clock — not a slow typewriter; no catch-up bursts |
|
|
| `Still` | no | static chevron | state-change redraws; stream still coalesces on the display clock |
|
|
|
|
Provider SSE deltas are **input**, never animation timing.
|
|
`StreamDisplayClock` (`tui/streaming`) coalesces them; `FrameRequester`
|
|
coalesces decorative frame wakes. The main `ui` poll loop remains the only
|
|
`terminal.draw` emitter — do not add a competing animation loop.
|
|
|
|
## Integration
|
|
|
|
- Derive `MotionPolicy::from_settings(low_motion, fancy_animations, force_reduced)`.
|
|
- Spinners: prefer `MotionPolicy::spinner_glyph` / `spinner_presentation`; the
|
|
frame table stays in `tui/spinner.rs`.
|
|
- Streaming: `stream_display_clock.set_allow_catch_up(policy.allows_catch_up_bursts())`.
|
|
- Working/phase chrome above the composer (TUI-DOG-008) must stay truthful under
|
|
Reduced/Still — calm redraws, not decorative spin.
|
|
|
|
## One-shot phase transitions
|
|
|
|
- A successful turn records the first history index owned by that turn. Tool
|
|
and agent receipts keep their final geometry and ordering while a bounded
|
|
70 ms stagger briefly dims then settles each row. Reduced/Still skip the
|
|
treatment and show the final receipts immediately.
|
|
- Ombre depth takes the typed `ShellPhase` as an input. Working leans subtly
|
|
deeper, verification leans toward the live surface ink, and waiting,
|
|
approval, and failure return the exact static base ramp.
|
|
- When an empty-water shell enters Working, fish follow one deterministic
|
|
800 ms flee-and-return arc keyed to `turn_started_at`. It never loops;
|
|
waiting, approval, stopped/error, and reduced-motion states remain still.
|
|
- These treatments never add/remove transcript rows, change hitboxes, or use
|
|
provider delta timing as an animation clock.
|
|
|
|
## Honesty note: catch-up is staged, not wired
|
|
|
|
`note_delta_with_backlog` and the catch-up thresholds exist and are tested,
|
|
but every production drain site currently calls `note_delta` (queued = 1), so
|
|
Full-motion catch-up never actually fires and Full/Reduced stream at the same
|
|
steady clock. Do not describe catch-up as live behavior until the real queue
|
|
depth/oldest-age metrics are fed in at the `ui.rs` drain sites
|
|
(TUI-DOG-017 follow-up).
|
|
|
|
A second, unrelated "adaptive chunking" policy (`streaming/chunking.rs`, plus a
|
|
`LineBuffer` newline gate) was deleted in v0.9.4: it could only ever decide
|
|
"drain everything available", and both `LineBuffer` constructors bypassed the
|
|
gate. A commit beat now unconditionally flushes everything received since the
|
|
previous beat. Newline-boundary safety for partial code fences is owned by the
|
|
incremental markdown parser (`ParseState::commit_complete_lines`), which is
|
|
where it is actually in force.
|