Skip to content

feat: syntax-highlighted code blocks + fuzzy / @-file prompt autocomplete - #33

Merged
saucam merged 2 commits into
mainfrom
feat/syntax-highlight-and-fuzzy-autocomplete
Jul 26, 2026
Merged

feat: syntax-highlighted code blocks + fuzzy / @-file prompt autocomplete#33
saucam merged 2 commits into
mainfrom
feat/syntax-highlight-and-fuzzy-autocomplete

Conversation

@saucam

@saucam saucam commented Jul 25, 2026

Copy link
Copy Markdown
Collaborator

What

Borrows three ideas from earendil-works/pi-tui — a mature TS TUI toolkit — adapted to our Ratatui stack. All three are additive, no architectural change.

1. Syntax-highlighted code blocks

Fenced code in assistant markdown is now highlighted by language via syntect (pure-Rust fancy-regex engine — no C toolchain / oniguruma, stays cross-compile friendly).

  • Common LLM fence tags (rust, python, ts, bash, yml, …) are normalised to a syntax syntect knows.
  • Unknown languages / parser hiccups fall back to the previous flat-green rendering.
  • Highlighting is stateful across a block (one HighlightLines per fence) and paid once per message version via the existing render cache — no per-frame cost.

2. @-file mention autocomplete

Typing @ in the prompt opens a filesystem palette (relative to the focused session's workdir), fuzzy-ranked against the partial path.

  • Tab completes the top hit: directories keep a trailing / so you can keep drilling; files get a trailing space.
  • Descends subdirectories (@src/re…), reveals dotfiles only when you type a leading dot.
  • Wired as a runtime-conditional Tab binding in app.rs (mirrors the existing esc-interrupts pattern) so the static keymap and all its tests stay untouched.

3. Fuzzy slash-command matching

The / palette and Tab completion now rank by a fuzzy subsequence score (consecutive-run + word-boundary bonuses, gap penalties) instead of strict prefix — /expsn finds /export, and Tab commits the best match.

New modules

  • fuzzy — subsequence scorer ported from pi-tui's fuzzy.ts
  • mention — filesystem suggestion ranking
  • render/highlightsyntect wrapper (lazy OnceLock syntax/theme sets)

Testing

  • cargo test --workspace green (314 tui tests + crate tests).
  • New unit tests across every new module; existing markdown/commands tests updated for the new behaviour.
  • cargo fmt --check clean; all new code is clippy-pedantic clean.

Notes

  • These were chosen from a broader comparison against pi-tui. Deliberately not borrowed: pi's differential renderer (Ratatui already double-buffers + we have O(log N) scrollback virtualization) and inline-scrollback model.
  • Follow-up PR (Set up crates.io + binary releases: publish-on-tag workflow #2 of 2): editor undo/redo + prompt history.

🤖 Generated with Claude Code

…lete

Borrows three ideas from earendil-works/pi-tui, adapted to our Ratatui stack:

- Syntax highlighting: fenced code in assistant markdown is highlighted by
  language via syntect (pure-Rust fancy-regex, no C toolchain). Common LLM
  fence tags are normalised to a known syntax; unknown languages fall back to
  the previous flat-green rendering. Highlighting is stateful across a block
  and paid once per message version through the existing render cache.

- @-file mention autocomplete: typing @ in the prompt opens a filesystem
  palette (relative to the focused session workdir), fuzzy-ranked against the
  partial path. Tab completes the top hit — directories keep a trailing slash
  to drill in, files get a trailing space. Handled as a runtime-conditional
  Tab binding, mirroring the existing esc-interrupts pattern.

- Fuzzy slash-command matching: the / palette and Tab completion now rank by
  a fuzzy subsequence score (consecutive-run + word-boundary bonuses, gap
  penalties) instead of strict prefix; Tab commits the best match.

New modules: fuzzy (scoring), mention (fs suggestions), render/highlight
(syntect wrapper). All new code is clippy-pedantic clean, with unit tests
across every new module.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@gemini-code-assist

Copy link
Copy Markdown

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

@socket-security

socket-security Bot commented Jul 25, 2026

Copy link
Copy Markdown

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff Package Supply Chain
Security
Vulnerability Quality Maintenance License
Addedsyntect@​5.3.010010093100100

View full report

* feat: prompt history recall + undo/redo

Second of two PRs borrowing prompt-editor ideas from pi-tui.

- Prompt history: Up/Down recall previously submitted prompts, shell-style.
  Recall is gated to the editor's vertical edges (Up on the first row, Down on
  the last) so multi-line cursor movement is untouched; it's wired as a
  runtime-conditional binding in the reducer (like the existing esc-interrupt
  and @-mention Tab) since it depends on cursor row + history state. Stepping
  down past the newest entry restores the stashed live draft. History is
  de-duplicated against the newest entry and capped at PROMPT_HISTORY_MAX.

- Undo/redo: Ctrl+Z / Ctrl+Y map to tui-textarea's built-in edit history via
  explicit keymap actions (discoverable + testable rather than relying on the
  library's default shortcuts).

State gains prompt_history / history_index / history_draft plus record_history,
set_prompt_text, history_prev, history_next. Full suite green; new unit tests
cover history walk/restore/dedup/cap and the new keymap bindings.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: drop redundant clone in prompt history recall (oracle review)

history_prev/next cloned prompt_history[i] into a String only to satisfy the
borrow checker (can't hold &self.prompt_history[i] across a &mut self method).
Replace the set_prompt_text method with a free `prompt_from_text(&str)` that
doesn't borrow self, so callers pass &self.prompt_history[i] directly — no
intermediate allocation. Behaviour unchanged; history tests still green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* refactor: back prompt history with VecDeque (oracle review)

Front-trimming a Vec on each cap is an O(n) shift; a VecDeque makes dropping
the oldest entry an O(1) pop_front, which is the semantically correct
structure for a bounded ring buffer with front removals. Negligible at the
200-entry cap, but cleaner and future-proof. Behaviour and tests unchanged.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
@saucam
saucam merged commit 64c2cfc into main Jul 26, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants