Summary

When the caret sits mid-line (real characters follow it before the next line break), inline ghost text has no home: it would render on top of the text after the caret. This promotes any such presentation to the popup (mirror) card instead, anchored to the caret line. It is the first step toward fill-in-middle (FIM) completions, which structurally need the card because a mid-line completion cannot be drawn inline.

The signal is the existing CaretLinePosition.isAtEndOfLine, so an end-of-line caret that still has later paragraphs below it stays inline (only same-line trailing characters trigger the card).

Validation

swiftlint lint --quiet <6 changed source files>
# exit 0

xcodebuild -project Cotabby.xcodeproj -scheme Cotabby -destination 'platform=macOS' \
  build-for-testing -derivedDataPath build/DerivedData CODE_SIGNING_ALLOWED=NO CODE_SIGNING_REQUIRED=NO
# ** TEST BUILD SUCCEEDED **

xcodebuild test-without-building ... \
  -only-testing:CotabbyTests/CompletionRenderModePolicyTests \
  -only-testing:CotabbyTests/CaretLinePositionTests \
  -only-testing:CotabbyTests/MirrorOverlayLayoutTests
# Executed 39 tests, with 0 failures

App-hosted tests were run locally with signing disabled. The 7 new policy tests cover the auto / always-inline / always-mirror / per-app-override matrix at both end-of-line and mid-line carets; existing policy, caret-position, and mirror-layout suites still pass.

Linked issues

No issue number provided. Groundwork for FIM completions.

Risk / rollout notes

• Behavior change to an existing flow: a suggestion generated with the caret mid-line (e.g. the mid-word continuations that already fire via MidWordContinuationPolicy) previously rendered as inline ghost text overlapping the trailing characters. It now renders in the popup card. This is a net improvement for that case, but it is a visible change.
• The mid-line rule deliberately overrides an explicit "Inline" appearance preference, because inline cannot render mid-line without overlap. If we would rather honor the pin and leave those users on inline mid-line, the override is a one-line change in CompletionRenderModePolicy.mode (gate the promotion on .auto only). Flagging for a product call.
• No schema, settings, or project.yml / pbxproj migrations. No new files (test changes live in existing files, so XcodeGen does not drift).
• Mirror-card anchoring for the new .caretMidLine reason reuses the trustworthy-geometry path (anchors to the caret line), so there is no new positioning math.

Greptile Summary

When the caret sits mid-line (real characters follow it on the same line), inline ghost text can't render without painting over those trailing characters. This PR detects that condition via the existing CaretLinePosition.isAtEndOfLine signal, promotes any inline result to the popup card with a new caretMidLine mirror reason, and anchors the card to the caret rect (geometry is trustworthy here) rather than the input-field rect.

• CompletionRenderModePolicy is split into a public mode(for:bundleIdentifier:) wrapper and a private preferenceMode helper; the wrapper applies the mid-line promotion as a single, self-contained rule that upgrades .inline only, leaving existing mirror reasons (e.g. .caretGeometryEstimated) unchanged.
• SuggestionOverlayGeometry gains isCaretAtEndOfLine: Bool (defaults to true for backward compatibility), threaded through SuggestionCoordinator+Acceptance and the withCaretRect copy helper.
• MirrorOverlayLayout.computeAnchorTopY groups .caretMidLine with .userPreference/.perAppOverride for caret-rect anchoring; seven new policy tests cover the full auto/always-inline/always-mirror/per-app matrix at both caret positions.

Confidence Score: 4/5

Safe to merge; the change is well-scoped, fully tested, and the backward-compatible default on isCaretAtEndOfLine preserves all existing call sites.

The logic is clean and the 7 new tests cover the full preference × caret-position matrix. The one doc comment in CompletionRenderMode.swift describes the wrong anchor rect for .caretMidLine (says field rect, implementation uses caret rect), which could mislead a future author working on FIM layout. No runtime behavior is affected.

Cotabby/Models/CompletionRenderMode.swift — the caretMidLine case doc needs the anchor-rect description corrected before FIM work builds on it.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[presentOverlay called] --> B[Build SuggestionOverlayGeometry\nwith isCaretAtEndOfLine from context]
    B --> C[CompletionRenderModePolicy.mode]
    C --> D[preferenceMode: check userPreference\n& per-app overrides]
    D --> E{effectivePreference}
    E -->|alwaysInline| F[.inline]
    E -->|alwaysMirror| G[.mirror reason: userPreference\nor perAppOverride]
    E -->|auto| H{caretQuality == .estimated?}
    H -->|yes| I[.mirror reason: caretGeometryEstimated]
    H -->|no| F
    F --> J{baseMode == .inline\nAND !isCaretAtEndOfLine?}
    G --> K[Return baseMode unchanged]
    I --> K
    J -->|yes| L[.mirror reason: caretMidLine]
    J -->|no| M[.inline]
    L --> N[MirrorOverlayLayout: anchor to caretRect\nfallback to inputFrameRect]
    M --> O[Inline ghost text at caret]
    K --> P{reason?}
    P -->|caretGeometryEstimated| Q[MirrorOverlayLayout: anchor to inputFrameRect\nfallback to caretRect offset]
    P -->|userPreference / perAppOverride / caretMidLine| N

_{Reviews (1): Last reviewed commit: "feat(overlay): show popup card when care..." | Re-trigger Greptile}

Greptile also left 1 inline comment on this PR.