Add AcceptanceGranularity setting (Word / Phrase / Full) for primary accept key

[rpickmans]

Summary

Adds a user-selectable acceptance granularity to the primary accept key (default Tab): Word (current behavior, default), Phrase (stops at .!?\n or their quoted/bracketed equivalents — Cursor-style), or Full Suggestion (same outcome as the dedicated full-accept key). The dedicated full-accept key (default `) remains an unconditional per-press override regardless of the chosen granularity.

The new nextAcceptancePhrase chunker composes over the existing nextAcceptanceChunk, so word-boundary, internal-punctuation, and leading-whitespace policy stay identical across multi-word seams. Two extra rules make sentence boundaries robust:

• Newlines are scanned inside each composed chunk — nextAcceptanceChunk returns leading whitespace including \n, so a tail like hello\nworld would otherwise carry the newline forward unnoticed.
• Terminator detection walks back past closing quotes and brackets, so "done." Next stops at the closing quote rather than over-accepting into the next sentence. Handles ", ', ), ], }, and curly quotes ”, ’.

Known limitation: a tail ending in U.S.A. is treated as a sentence end (early break). Rule-based scanners can't disambiguate this without NLP; Cursor and Copilot behave the same way. Covered by a dedicated test that documents the behavior.

Validation

xcodebuild -project Cotabby.xcodeproj -scheme Cotabby \
  -destination 'platform=macOS' CODE_SIGNING_ALLOWED=NO test
# ** TEST SUCCEEDED **  Executed 449 tests, with 1 test skipped and 0 failures

swiftlint lint --quiet
# exit 0

18 new nextAcceptancePhrase tests cover empty tail, no-terminator, .!? boundaries, newline-between-tokens, leading-newline, multiple newlines, leading whitespace, abbreviation false-break (documented as known limitation), autoAcceptTrailingPunctuation flag invariance, interior punctuation preservation, straight/curly quoted sentence ends, parenthesized sentence ends, nested closers, and bare-closing-quote (no false break).

Linked issues

Fixes #10

Risk / rollout notes

• Default unchanged. New installs and existing installs without the persisted UserDefaults key get .word, preserving current Tab behavior. The setting is read once per accept call from the value-typed snapshot, so a mid-press toggle in Settings can't strand a partially-handled press.
• Settings exposed in both UIs. Picker added to both the legacy SettingsView.shortcutsSection and the redesigned ShortcutsPaneView so the setting is reachable regardless of which Settings shell is active.
• UI naming question for maintainer: the "Accept Word" row label was kept as-is to stay symmetric with the WelcomeView onboarding (WelcomeView.swift:303) and the parallel label in the other settings pane. The new "Acceptance Mode" Picker disambiguates what the key actually does. If you'd prefer a dynamic row label ("Accept Word" / "Accept Phrase" / "Accept Full") to match the picker selection, happy to make that change in this PR or a follow-up.
• Per-app and per-profile granularity stay out of scope per the design discussion on the issue — those touch the rules system and warrant a separate PR.
• prepareAcceptance signature change (added required granularity: parameter, no default) — only 2 in-tree callers (coordinator + state-helper test); both updated. .full is precondition-rejected in prepareAcceptance because the coordinator routes full accepts to prepareFullAcceptance for the distinct invalid-reason message; defensive handling could be substituted if you'd prefer.

Greptile Summary

This PR adds a user-selectable AcceptanceGranularity setting (Word / Phrase / Full) for the primary accept key, while keeping the dedicated full-accept key as an unconditional per-press override. The new nextAcceptancePhrase chunker composes over nextAcceptanceChunk, inheriting existing word-boundary and leading-whitespace policy, with extra rules for newlines (which appear as leading whitespace in the composed chunks) and closing-punctuation walk-back for quoted/parenthesised sentence ends.

• AcceptanceGranularity is persisted in UserDefaults with a .word fallback, wired into the settings snapshot publisher via an outer CombineLatest, and exposed in both the legacy SettingsView and the redesigned ShortcutsPaneView through a new shared AcceptanceModePickerView component.
• The prepareAcceptance signature gains a required granularity: parameter; .full is guarded with assertionFailure + graceful fallback since the coordinator already routes full accepts to prepareFullAcceptance. 18 new unit tests cover the chunker in isolation; a new integration test in SuggestionStateHelperTests confirms the .phrase branch propagates correctly through prepareAcceptance end-to-end.

Confidence Score: 5/5

Safe to merge — the new phrase chunker composes cleanly over the existing word chunker, default behavior is unchanged for existing installs, and all previously flagged issues have been addressed.

The core logic in nextAcceptancePhrase is well-reasoned: newline detection correctly captures
from leading-whitespace chunks, the closing-punctuation walk-back handles nested quoted/parenthesised sentence ends, and autoAcceptTrailingPunctuation invariance holds. The .full guard now uses assertionFailure + graceful .invalid return instead of preconditionFailure, matching every other invalid branch. The integration test for .phrase in prepareAcceptance closes the only gap from the prior review round. No new behavioral regressions are introduced; the .word default preserves all existing acceptance behaviour for users who never touch the new setting.

No files require special attention.

Important Files Changed

Filename	Overview
Cotabby/Support/SuggestionSessionReconciler.swift	Adds nextAcceptancePhrase and endsInSentenceTerminator; logic is sound — newline scan correctly captures \n from leading-whitespace chunks, closing-punc walk-back handles nested/quoted sentence ends, and autoAcceptTrailingPunctuation invariance is verified by tests.
Cotabby/Services/Suggestion/SuggestionInteractionState.swift	prepareAcceptance gains a required granularity: parameter; .full now uses assertionFailure + .invalid return instead of preconditionFailure, matching the graceful degradation pattern of every other invalid branch.
Cotabby/App/Coordinators/SuggestionCoordinator+Acceptance.swift	Correctly reads granularity from the settings snapshot and routes .full (and the fullText flag) to prepareFullAcceptance so .full never reaches prepareAcceptance.
Cotabby/Models/SuggestionSettingsModel.swift	Persists acceptanceGranularity via UserDefaults rawValue with a safe .word fallback; adds setAcceptanceGranularity with early-exit guard; layers $acceptanceGranularity into the snapshot publisher via a second CombineLatest around the existing CombineLatest4.
Cotabby/UI/Settings/Components/AcceptanceModePickerView.swift	New shared picker component that eliminates the verbatim duplication between SettingsView and ShortcutsPaneView; clean Binding wrapping via setAcceptanceGranularity.
CotabbyTests/SuggestionSessionReconcilerTests.swift	18 new tests cover the phrase chunker thoroughly: empty input, no terminator, .!? stops, newline variants, leading whitespace, abbreviation false-break (documented known limitation), flag invariance, interior punctuation, straight/curly quotes, parentheses, nested closers, and bare closing quote non-break.
CotabbyTests/SuggestionStateHelperTests.swift	Adds integration test for .phrase granularity through prepareAcceptance, confirming the accepted chunk spans to the sentence terminator and that session.remainingText is unmodified before commit.
Cotabby/Models/SuggestionEngineModels.swift	AcceptanceGranularity enum added as String/Codable/CaseIterable/Sendable; acceptanceGranularity field added to SuggestionSettingsSnapshot.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Key press detected] --> B{fullText flag\nor granularity == .full?}
    B -- Yes --> C[prepareFullAcceptance\nreturns entire remainingText]
    B -- No --> D{granularity}
    D -- .word --> E[nextAcceptanceChunk\none word + optional trailing punct]
    D -- .phrase --> F[nextAcceptancePhrase\ncompose chunks until .!? or newline]
    F --> G{chunk contains newline?}
    G -- Yes --> H[Accumulate up to newline inclusive\nreturn early]
    G -- No --> I[Accumulate chunk\nadvance working]
    I --> J{endsInSentenceTerminator?\nwalk back past closing punct}
    J -- Yes --> K[Return phrase]
    J -- No --> F
    E --> L[guard chunk not empty]
    K --> L
    H --> L
    L -- empty --> M[.invalid — key passes through]
    L -- non-empty --> N[.ready liveContext / session / acceptedChunk]
    N --> O[Coordinator calls commitAcceptedChunk]

Loading

Comments Outside Diff (1)

1. Cotabby/UI/Settings/Panes/ShortcutsPaneView.swift, line 381-386 (link)

   Verbatim Picker duplication across both settings UIs

   The five-line Picker block and its acceptanceGranularityBinding helper are replicated character-for-character in SettingsView.swift. If the labels or cases change (e.g., renaming "Full Suggestion" or adding a fourth mode), both files need identical edits. Extracting this into a small private AcceptanceModePickerView or a @ViewBuilder helper shared by both would remove the drift risk entirely.

   Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

   

_{Reviews (2): Last reviewed commit: "Harden AcceptanceGranularity: graceful ...." | Re-trigger Greptile}

[rpickmans]

Pushed 24233cf addressing the three points:

• The .full arm of prepareAcceptance now uses assertionFailure plus an .invalid return, so a stray .full degrades gracefully in release.
• Pulled the duplicated Acceptance Mode picker and its binding out of SettingsView and ShortcutsPaneView into a shared AcceptanceModePickerView, so the labels and cases live in one place.
• Added the .phrase integration test through prepareAcceptance.

Build and the full test suite pass (450 tests), SwiftLint is clean, and the project file was regenerated with xcodegen.

[FuJacob]

Thanks @rpickmans will take a look tonight! :)

[FuJacob]

LGTM thank you for your hard work!! :)