dotfiles/.config/dot/.claude/tasks/0009-kde-shortcut-completion.md

blocked-by

blocked-by
0005-kde-shortcuts-mechanism

What to build

dot kde save's tab-completion (cmd_complete in commands/kde/kde.py) currently only enumerates schema-backed identifiers via iter_schema_identifiers — it was built as a side effect of the diff task (0003) and never revisited when the shortcuts mechanism (0005) landed. Extend cmd_complete to also enumerate shortcut identifiers.

Source the shortcut identifiers live via kglobalaccel, mirroring how schema identifiers are freshly parsed from .kcfg files on every call: call allMainComponents() to get every registered component's componentUnique, then allActionsForComponent() per component (already used by _resolve_shortcut_action_id) to get every actionUnique, yielding kglobalshortcutsrc.<componentUnique>.<actionUnique> candidates. No caching — walk fresh on every invocation.

Print shortcut identifiers as their own block, after the existing schema-backed block — not merged into one interleaved sorted list. Keep them plain, with no friendly-name description text, matching the existing schema-identifier output style.

If the D-Bus walk fails for any reason — a non-zero busctl exit (RuntimeError, already raised by _kglobalaccel_call) or busctl itself being missing (OSError from subprocess.run) — swallow it silently: omit the shortcuts block, still print the schema block, and emit no stderr diagnostic.

Freeform identifiers (e.g. kxkbrc.Layout.Options) are explicitly out of scope for this task: there is no schema to enumerate them from, so this stays a permanent, accepted completion gap, not something to fix here.

Acceptance criteria

• python3 kde.py complete includes every currently-registered kglobalshortcutsrc.<componentUnique>.<actionUnique> identifier, sourced live via allMainComponents/allActionsForComponent
• Schema-backed identifiers print first, followed by shortcut identifiers, as two distinct blocks — not interleaved into one merged sorted list
• Shortcut identifiers print plain, with no friendly-name description text
• If the D-Bus walk raises RuntimeError or OSError, the shortcuts block is omitted, the schema block still prints normally, and nothing is written to stderr
• Freeform identifiers remain unlisted by cmd_complete (unchanged, confirmed not a regression)
• Verified manually against a live session — no new automated tests, consistent with the existing shortcuts-mechanism test carve-out (spec's testing decisions, 0005's Implementation Notes)

Implementation Notes

• iter_shortcut_identifiers (new, commands/kde/kde.py) walks allMainComponents() then allActionsForComponent() per component, yielding kglobalshortcutsrc.<componentUnique>.<actionUnique>. cmd_complete wraps that walk in sorted(set(...)) and appends it as a second print loop after the existing schema-backed one, inside a try/except (RuntimeError, OSError) that falls back to an empty list on any failure — so a missing busctl or an unreachable D-Bus session degrades completion instead of breaking it.
• Manually verified both paths: live run on this machine prints 278 shortcut identifiers after 322 schema-backed ones; with busctl removed from PATH (simulating a non-KDE/minimal shell), cmd_complete still exits 0, prints only the 322 schema identifiers, and writes nothing to stderr.
• /review-uncommitted's Standards pass flagged two judgement-call smells: (1) the D-Bus call/unpack idiom for allActionsForComponent was duplicated between the new function and _resolve_shortcut_action_id; (2) the silent except swallow had no comment explaining why. Fixed both: extracted a shared _actions_for_component(component_unique) helper used by both call sites, and added a comment on the try explaining that fish invokes this on every TAB press in shells that may lack a live KDE session, so a broken shortcuts source must never cost the already-printed schema candidates. Re-ran the full test suite (101/101 pass) and both manual checks after the fix.
• No automated tests added, per the task's own acceptance criterion and the shortcuts mechanism's existing test carve-out (0005's Implementation Notes: a live D-Bus session isn't practically substitutable without disproportionate mock infrastructure).