| |
|
|
|
|
| |
The bug fixes raised line coverage on duet.el to 92.3%. This closes the rest. The new tests exercise the branches a review would want locked down before execution and UI build on this core: each arm of the minimal failure normalizer (missing executable, stall, signal cancellation, unknown-without-exit), a :match predicate rather than a regexp, the rsync normalizer mapping known stderr to permission/space/protocol classes, the capability-tier contract check, an explicit :async override, the case-collision path through the safety composite's injected predicates, and redaction of a pattern with no capture group.
Line coverage on duet.el is now 100%. The one deliberately trivial case left is the entry command, a not-yet-implemented stub that errors. Its test asserts that and will change when the pane layout lands.
|
|
|
I brought the skeleton up to a working package baseline (Phase 0 in the design spec). Eask defines the package and its dev deps. A root Makefile delegates test targets to tests/Makefile and adds compile, coverage, lint, doctor, and clean, matching the layout the other packages use.
deps installs both halves DUET needs: the Emacs Lisp deps via eask, and the transport CLIs (rsync, rclone, lftp, unison) via the system package manager, so a contributor's environment is ready before the code that shells out to them.
make complexity runs a small homegrown McCabe branch counter (scripts/duet-complexity.el). No off-the-shelf tool measures Emacs Lisp: lizard doesn't support it and codemetrics is an interactive overlay, so DUET owns one. The counting is pure and covered by Normal/Boundary/Error tests. The budget is soft and the target is advisory.
The ERT harness (bootstrap, check-deps, per-file undercover coverage) and a smoke test prove the loop works end to end.
|
