Terminal ghost suggetions

[gajendraxdev]

Summary by CodeRabbit

• Bug Fixes

  • Improved Ghost suggestion spacing and normalization for new path arguments, mid-token completions, and path fragments.
  • Enhanced suggestion ranking so mid-token completions can correctly override higher-frecency spaced history.
  • Fixed Ghost overlay positioning by anchoring to the first-caret origin and compensating for SSH/local echo lag.
  • Improved terminal rendering/segmentation for wide characters, emoji (ZWJ), and combining marks; prevented duplicate or missing spaces.

• Release

  • Updated version to 2.22.1 and refreshed the changelog.

• Tests

  • Added/expanded unit coverage for spacing behavior, ranking, and ghost cursor/width calculations.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 04f86f80-2f22-44ae-80bb-f6cb3d6e551c

📥 Commits

Reviewing files that changed from the base of the PR and between bb613d2 and 64fb421.

📒 Files selected for processing (4)

• src-tauri/src/ghost/path_suggest.rs
• src-tauri/src/ghost/ranking.rs
• src/lib/ghostSuggestions/displayWidth.ts
• tests/ghostSuggestionsHelpers.test.mjs

---

📝 Walkthrough

Walkthrough

Ghost suggestion spacing and ranking now favor provider-defined boundaries and mid-token continuations. The terminal overlay gains SSH-lag-aware positioning, display-width segmentation, layout propagation, updated tests, and 2.22.1 release metadata.

Changes

Ghost Suggestions

Layer / File(s)	Summary
Backend spacing and ranking src-tauri/src/ghost/{suffix.rs,ranking.rs,path_suggest.rs}, src/lib/ghostSuggestions/{suggestionSuffix.ts,client.ts}, tests/ghostSuggestionsHelpers.test.mjs	Suffix normalization preserves provider spacing, path completions identify new arguments, and ranking prefers mid-token continuations with regression coverage.
Ghost position and cell measurement src/lib/ghostSuggestions/{cursorPosition.ts,displayWidth.ts}, tests/ghostSuggestionsHelpers.test.mjs, tsconfig.agent-tests.json	Ghost layout contracts, lag-aware caret resolution, terminal display-width utilities, segmentation, and helper tests are added.
Layout capture and overlay rendering src/components/terminal/{useTerminalGhost.ts,Terminal.tsx,TerminalHost.tsx,GhostSuggestionOverlay.tsx}	Typed-line origin and width flow through the terminal components, while the overlay tracks echo events and renders cell-sized suffix segments.
Release metadata package.json, src-tauri/Cargo.toml, src-tauri/tauri.conf.json, CHANGELOG.md	Version references and changelog links are updated for release 2.22.1.

Estimated code review effort: 4 (Complex) | ~60 minutes

Sequence Diagram(s)

sequenceDiagram
  participant TerminalInput
  participant useTerminalGhost
  participant TerminalHost
  participant GhostSuggestionOverlay
  participant getGhostPixelPosition
  TerminalInput->>useTerminalGhost: process typed input and echo events
  useTerminalGhost->>useTerminalGhost: capture origin and typed cell count
  useTerminalGhost->>TerminalHost: provide ghostLayout
  TerminalHost->>GhostSuggestionOverlay: pass layout and suggestion
  GhostSuggestionOverlay->>getGhostPixelPosition: resolve ghost pixel position
  getGhostPixelPosition-->>GhostSuggestionOverlay: return predicted or live position

Loading

Possibly related PRs

• zync-sh/zync#60: Introduced the related ghost suggestion Rust and frontend implementation.
• zync-sh/zync#79: Updated the shared ghost cursor-positioning logic.
• zync-sh/zync#83: Modified the related ghost suffix, ranking, and path suggestion behavior.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title is concise and clearly points to the main theme of the change: terminal ghost suggestions.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[qodo-code-review]

PR Summary by Qodo

Fix ghost suggestion spacing and SSH echo-lag overlay alignment

🐞 Bug fix 🧪 Tests 📝 Documentation ⚙️ Configuration changes 🕐 40+ Minutes

AI Description

• Stop ghost suffix normalization from inventing leading spaces; providers own word boundaries.
• Prefer mid-token history continuations in ranking to avoid ls → ls blk regressions.
• Reposition and size ghost overlay using predicted typed-cell layout and terminal cell metrics.

Diagram

graph TD
  U["User typing"] --> UTG["useTerminalGhost"] --> IT["InputTracker"] --> GL["GhostLayoutHint"] --> CP["cursorPosition"] --> GSO["GhostSuggestionOverlay"] --> XT["xterm.js"]
  UTG --> CL["ghost client"] --> BK["Rust ghost engine"]
  BK --> RK["ranking + suffix"] --> CL
  BK --> PS["path_suggest"] --> CL
  DW["displayWidth helpers"] --> UTG
  DW --> GSO
  subgraph Legend
    direction LR
    _ui["UI/React"] ~~~ _lib["TS helpers"] ~~~ _be["Rust backend"]
  end

Loading

High-Level Assessment

The following are alternative approaches to this PR:

1. Use xterm's internal unicode/wcwidth service

• ➕ Better fidelity with xterm rendering rules across edge-case Unicode sequences
• ➕ Avoids maintaining a partial EastAsianWidth/combining-range table
• ➖ May rely on private/internal APIs that can change across xterm versions
• ➖ Harder to test outside of a real terminal instance

2. Adopt a dedicated wcwidth/grapheme library

• ➕ More complete Unicode coverage than a curated subset
• ➕ Potentially simpler to reason about and reuse
• ➖ Adds dependency weight (bundle size/audit surface)
• ➖ Still needs careful integration with terminal cell measurement and overlay spans

3. Keep frontend spacing heuristics (status quo) but tighten rules

• ➕ Less backend change; fewer provider contracts
• ➕ No need to coordinate path/history spacing ownership
• ➖ Heuristics are fragile and already caused ls blk-style corruption
• ➖ Harder to ensure consistent behavior across Rust/TS test layers

Recommendation: The PR’s approach (provider-owned spacing + explicit mid-token preference + lag-aware overlay layout) is the most robust: it eliminates a class of heuristic spacing bugs, makes history/path providers responsible for boundaries, and fixes SSH echo lag without needing to depend on xterm internals. If Unicode-width edge cases become frequent, consider switching the new display-width subset to a well-maintained wcwidth implementation or xterm-provided width logic.

Files changed (17) +579 / -242

Enhancement (4) +236 / -4

Terminal.tsxPlumb ghost layout hint through terminal component +2/-0

Plumb ghost layout hint through terminal component

• Exposes 'ghostLayout' from the ghost hook and passes it down to the terminal host so the overlay can use lag-aware positioning.

src/components/terminal/Terminal.tsx

TerminalHost.tsxPass ghost layout hint into GhostSuggestionOverlay +8/-1

Pass ghost layout hint into GhostSuggestionOverlay

• Extends host props to accept an optional 'ghostLayout' and forwards it to the overlay when inline ghost suggestions are enabled.

src/components/terminal/TerminalHost.tsx

cursorPosition.tsAdd lag-aware ghost cell resolution and wrapping helpers +105/-3

Add lag-aware ghost cell resolution and wrapping helpers

• Introduces layout hint types, terminal wrapping math, and a resolver that prefers predicted end-of-line when the live caret lags remote echo. Exposes 'getGhostPixelPosition' for overlays while keeping 'getCursorPixelPosition' as a live-caret wrapper.

src/lib/ghostSuggestions/cursorPosition.ts

displayWidth.tsAdd terminal display-width and segmentation utilities +121/-0

Add terminal display-width and segmentation utilities

• Adds a lightweight wcwidth-like implementation for zero-width, wide, and emoji ranges plus helpers to compute display-cell counts. Provides grapheme-aware segmentation (Intl.Segmenter when available) for cell-aligned overlay rendering.

src/lib/ghostSuggestions/displayWidth.ts

Bug fix (6) +258 / -228

path_suggest.rsMake path provider own leading space for new path args +14/-1

Make path provider own leading space for new path args

• Introduces a 'new_path_arg' flag to distinguish a new argument after a bare command vs a mid-token fragment. When completing a new path argument, the provider prepends a leading space to the suggested piece, preventing suffix normalization from inventing spaces for history mid-token completions.

src-tauri/src/ghost/path_suggest.rs

ranking.rsPrefer mid-token history continuations in suffix ranking +60/-19

Prefer mid-token history continuations in suffix ranking

• Adds mid-token detection and incorporates it into an effective score so command-name continuations outrank spaced new-word entries. Updates candidate sorting accordingly and adds a regression test ensuring 'ls' completes to 'lsblk' over 'ls blk' even with higher frecency.

src-tauri/src/ghost/ranking.rs

suffix.rsStop normalizer from inventing spaces; treat spacing as provider-owned +48/-139

Stop normalizer from inventing spaces; treat spacing as provider-owned

• Simplifies suffix normalization to only strip duplicate leading whitespace when the typed line already ends with whitespace. Removes heuristic glue/space insertion logic and updates unit tests to reflect provider-owned boundaries (history/path).

src-tauri/src/ghost/suffix.rs

GhostSuggestionOverlay.tsxRender ghost suffix at predicted typed end and align spans to cell width +50/-9

Render ghost suffix at predicted typed end and align spans to cell width

• Switches overlay positioning from live cursor-only to lag-aware ghost positioning using a layout hint. Adds remeasure triggers on both input and terminal writes, and renders the suggestion via per-cell segments sized with measured xterm cell width to prevent drift with wide/combining glyphs.

src/components/terminal/GhostSuggestionOverlay.tsx

useTerminalGhost.tsTrack typed-line origin and display-cell count for SSH lag compensation +81/-4

Track typed-line origin and display-cell count for SSH lag compensation

• Adds 'ghostLayout' state capturing the first typed cell origin and the display-cell width of the locally typed line buffer. Updates event handling to seed origin before feed, refresh layout on suggestion/accept/clear, and reset layout on disconnect/disable.

src/components/terminal/useTerminalGhost.ts

suggestionSuffix.tsMirror Rust normalizer: never add leading space in TS helper +5/-56

Mirror Rust normalizer: never add leading space in TS helper

• Removes the frontend heuristic spacing logic so the TS mirror matches the simplified Rust behavior: only strip duplicate leading whitespace after trailing whitespace; otherwise pass suffix through.

src/lib/ghostSuggestions/suggestionSuffix.ts

Tests (1) +67 / -6

ghostSuggestionsHelpers.test.mjsAdd tests for display width, segmentation, and lag-aware positioning +67/-6

Add tests for display width, segmentation, and lag-aware positioning

• Extends agent tests to cover display-cell width calculations, segmentation behavior for wide glyphs, the updated spacing invariant (never invent leading spaces), and the ghost cell prediction logic for SSH echo lag and wrapping/clamping.

tests/ghostSuggestionsHelpers.test.mjs

Documentation (2) +13 / -1

CHANGELOG.mdDocument ghost spacing, lag overlay, and cell-metric fixes +12/-1

Document ghost spacing, lag overlay, and cell-metric fixes

• Adds a 2.22.1 release entry detailing fixes for mid-token spacing, SSH echo-lag overlay placement, and terminal display-width alignment. Updates compare links for the new tag.

CHANGELOG.md

client.tsClarify backend ownership of suffix spacing +1/-0

Clarify backend ownership of suffix spacing

• Documents that the backend owns spacing/ranking and the frontend should not rewrite suffixes.

src/lib/ghostSuggestions/client.ts

Other (4) +5 / -3

package.jsonBump app version to 2.22.1 +1/-1

Bump app version to 2.22.1

• Updates the JS package version to 2.22.1 to match the release.

package.json

Cargo.tomlBump Tauri/Rust package version to 2.22.1 +1/-1

Bump Tauri/Rust package version to 2.22.1

• Updates the Rust crate version to 2.22.1 for the release.

src-tauri/Cargo.toml

tauri.conf.jsonBump Tauri config version to 2.22.1 +1/-1

Bump Tauri config version to 2.22.1

• Updates the Tauri app version field to 2.22.1.

src-tauri/tauri.conf.json

tsconfig.agent-tests.jsonInclude new cursor/layout and display-width helpers in agent test build +2/-0

Include new cursor/layout and display-width helpers in agent test build

• Adds the new TS helper modules to the agent test compilation set.

tsconfig.agent-tests.json

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0) 🎨 UX issues (0) 🔗 Cross-repo conflicts (0) 📜 Skill insights (0)

1. Wrong cd cd completion ✓ Resolved 🐞 Bug ≡ Correctness

Description

In path_suggestions, any directory-only command whose last arg equals the command name is treated
as a bare command, clearing partial and setting new_path_arg. This breaks completing paths like
cd cd (directory named cd) and can insert an extra leading space into the suggested suffix.

Code

src-tauri/src/ghost/path_suggest.rs[R374-380]

        partial = last_arg.clone();
        if is_directory_only && partial.to_ascii_lowercase() == command_name.to_ascii_lowercase() {
            partial.clear();
+            new_path_arg = true;
+        } else if last_arg.is_empty() {
+            new_path_arg = true;
        }

Relevance

⭐⭐⭐ High

Team heavily iterates on ghost path spacing/correctness; similar ghost path reliability fixes
accepted in PR #83.

PR-#83

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

get_last_arg() returns the last whitespace-delimited token, so for cd cd it returns cd, which
equals the command name; the new directory-only branch clears partial and marks new_path_arg,
which then triggers space-prefixing in the output builder.

src-tauri/src/ghost/token.rs[66-97]
src-tauri/src/ghost/path_suggest.rs[374-381]
src-tauri/src/ghost/path_suggest.rs[577-582]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
`path_suggestions()` treats `last_arg == command_name` as a bare directory-only command invocation and clears `partial`, which is incorrect when the user actually typed an argument equal to the command name (e.g. `cd cd`). This also sets `new_path_arg`, causing the path provider to prepend a space and generating an invalid completion.

## Issue Context
- `get_last_arg()` returns the substring after the last unquoted space; for `"cd cd"` it returns `"cd"`, matching `get_command_name()`.
- The current heuristic doesn’t verify that the command truly has *no* arguments (token count == 1 / ends-with-space handling).

## Fix Focus Areas
- src-tauri/src/ghost/path_suggest.rs[374-381]
- src-tauri/src/ghost/path_suggest.rs[577-582]

## Suggested fix
- Only enter the `partial.clear(); new_path_arg = true;` path when the input is truly a bare directory-only command (e.g. token count is 1, or the line has no whitespace after the command).
 - Example: compute a token count from `parse_line.trim_end()` (quote-aware if needed), and require `token_count == 1` before treating `last_arg == command_name` as “bare”.
- Add a regression test for `cd cd` (and optionally `pushd pushd` / `popd popd`) verifying that `partial` remains `"cd"` and suggestions are filtered/prefixed correctly (no injected leading space unless starting a new arg).

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

2. Combining marks become invisible ✓ Resolved 🐞 Bug ☼ Reliability

Description

When Intl.Segmenter is unavailable, segmentTerminalCells() falls back to per-code-point
segmentation, producing standalone combining-mark segments with cells = 0.
GhostSuggestionOverlay renders these as 0px-wide spans with overflow: hidden, making the marks
disappear in the ghost overlay.

Code

src/components/terminal/GhostSuggestionOverlay.tsx[R125-142]

+      {segments && cellWidth
+        ? segments.map((seg, i) => (
+            <span
+              key={`${i}:${seg.text}`}
+              style={{
+                display: 'inline-block',
+                // Wide CJK/emoji → 2 cells; combining-only clusters → skip width 0
+                width: Math.max(seg.cells, 0) * cellWidth,
+                minWidth: seg.cells > 0 ? undefined : 0,
+                height: cellHeight,
+                lineHeight: `${cellHeight}px`,
+                textAlign: 'center',
+                overflow: 'hidden',
+              }}
+            >
+              {seg.text}
+            </span>
+          ))

Relevance

⭐⭐⭐ High

They prioritize ghost overlay width accuracy (wide/combining marks); recent ghost overlay/layout
fixes accepted in PR #83.

PR-#83

ⓘ Recommendations generated based on similar findings in past PRs

Evidence

The fallback segmentation can emit standalone code points; combining marks are classified as
zero-width, yielding segments with cells=0, and the overlay then renders those segments in
0-width, clipped spans.

src/lib/ghostSuggestions/displayWidth.ts[7-72]
src/lib/ghostSuggestions/displayWidth.ts[85-121]
src/components/terminal/GhostSuggestionOverlay.tsx[125-142]

Agent prompt

The issue below was found during a code review. Follow the provided context and guidance below and implement a solution

## Issue description
In the no-`Intl.Segmenter` fallback, combining marks (width 0) can become standalone segments. The overlay renders each segment in its own `inline-block` with `width = seg.cells * cellWidth` and `overflow: hidden`, so any segment with `cells = 0` has zero render area and the mark becomes invisible.

## Issue Context
- `segmentTerminalCells()` explicitly supports a fallback segmentation path.
- `codePointDisplayWidth()` assigns many combining marks width 0.
- The overlay’s segmented rendering assumes segments with `cells == 0` are still safe to render as individual spans.

## Fix Focus Areas
- src/lib/ghostSuggestions/displayWidth.ts[85-121]
- src/components/terminal/GhostSuggestionOverlay.tsx[125-142]

## Suggested fix
- In `segmentTerminalCells()` fallback mode (and/or generally), merge any leading/standalone zero-width code points (combining marks, variation selectors, ZWJ sequences as appropriate) into the previous non-zero segment’s `text` instead of emitting a separate `{ cells: 0 }` segment.
 - This keeps `cells` driven by the base glyph while preserving the full grapheme text.
- Optionally harden the overlay rendering by avoiding `overflow: hidden` for `cells == 0` spans (or skipping standalone zero-cell spans entirely after merge).
- Add/extend agent tests to cover a string like `"e\u0301"` in a forced fallback path and assert that the segments preserve the combining mark without producing a separate zero-cell span.

ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools

---

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

src-tauri/src/ghost/ranking.rs (1)

71-77: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Extract the mid-token bonus into a named constant.

50.0 is a hard-coded value that fully dominates suffix_bonus_for_command's bonus range (including the -3 shell-separator and -2 chained-command penalties). That's likely intentional per the PR's stated goal, but as a bare magic number it's hard to tune/reason about later, and it effectively neutralizes the existing safety-oriented penalties for mid-token candidates.

♻️ Proposed refactor

+/// Mid-token continuations should outrank a "new word" match unless the
+/// combined structural bonus for the new-word candidate is enormous. Chosen
+/// to be larger than the full `suffix_bonus_for_command` range.
+const MID_TOKEN_BONUS: f64 = 50.0;
+
 fn effective_rank_score(frecency: f64, bonus: i32, mid_token: bool) -> f64 {
-    let mid = if mid_token { 50.0 } else { 0.0 };
+    let mid = if mid_token { MID_TOKEN_BONUS } else { 0.0 };
     frecency + mid + f64::from(bonus)
 }

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src-tauri/src/ghost/ranking.rs` around lines 71 - 77, Extract the hard-coded
50.0 mid-token score in effective_rank_score into a clearly named constant, then
use that constant when mid_token is true. Keep the existing scoring behavior and
suffix bonus handling unchanged.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/lib/ghostSuggestions/displayWidth.ts`:
- Around line 85-121: Update segmentTerminalCells to calculate each grapheme
cluster’s terminal width using grapheme-aware logic rather than relying solely
on stringDisplayWidth(segment). Ensure ZWJ emoji, skin-tone sequences, and
combining-mark clusters receive the correct cell count while preserving
zero-width standalone characters and the existing segment filtering behavior.

---

Nitpick comments:
In `@src-tauri/src/ghost/ranking.rs`:
- Around line 71-77: Extract the hard-coded 50.0 mid-token score in
effective_rank_score into a clearly named constant, then use that constant when
mid_token is true. Keep the existing scoring behavior and suffix bonus handling
unchanged.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 1b367e26-792d-4e71-8f85-43abf116d7e0

📥 Commits

Reviewing files that changed from the base of the PR and between 31b8893 and bb613d2.

⛔ Files ignored due to path filters (2)

• package-lock.json is excluded by !**/package-lock.json
• src-tauri/Cargo.lock is excluded by !**/*.lock

📒 Files selected for processing (17)

• CHANGELOG.md
• package.json
• src-tauri/Cargo.toml
• src-tauri/src/ghost/path_suggest.rs
• src-tauri/src/ghost/ranking.rs
• src-tauri/src/ghost/suffix.rs
• src-tauri/tauri.conf.json
• src/components/terminal/GhostSuggestionOverlay.tsx
• src/components/terminal/Terminal.tsx
• src/components/terminal/TerminalHost.tsx
• src/components/terminal/useTerminalGhost.ts
• src/lib/ghostSuggestions/client.ts
• src/lib/ghostSuggestions/cursorPosition.ts
• src/lib/ghostSuggestions/displayWidth.ts
• src/lib/ghostSuggestions/suggestionSuffix.ts
• tests/ghostSuggestionsHelpers.test.mjs
• tsconfig.agent-tests.json