The checks.smoke derivation only ran 'kitchen --help', which the e2e
tier now covers (end_to_end_test.ts, '--help lists the view command').
checks.tests is the remaining flake check.
Fold the view command module (viewFile, the error type, the read
adapter) into index.ts so the command functions sit beside the guarded
CLI, and generalise formatViewError into formatError matching over a
CliError union that new fallible commands extend.
Align the test files with ADR 0009's tiers by name:
- index_test.ts now holds the in-process integration tests of the
command functions (formerly view_test.ts)
- end_to_end_test.ts holds the black-box subprocess e2e (formerly
index_test.ts)
Guard the CLI entry's program.parse() behind import.meta.main so the
command module can be imported without running the CLI, then rebuild the
deleted suite across the three tiers ADR 0009 defines:
- unit: parse (core) and render (bin), asserted on return values
- integration: the view command function, asserted on its Result
- e2e: the kitchen binary as a subprocess (exit codes, streams, ANSI
suppression, errors, --help)
Drop the stale test-support.ts coverage ignore and refresh the task's
implementation notes to describe the tiered suite.
Add ADR 0009 defining the unit / integration / e2e tiers by the seam
each exercises, with the value-returning vs process-side-effecting line
as the boundary between in-process and e2e-only. Add one line to ADR
0008 noting the Result seam is what makes error propagation assertable
in-process, cross-linking 0009.
Temporary: clear the entire test suite (unit, subprocess, the shared
helper, and the parser-integration placeholder) so the tests can be
rebuilt from scratch. Fixtures and the test tooling (bunfig, report
script) are kept. Note: `bun test` now errors on an empty suite, so the
flake's tests check is red until tests return.
Split the view command out of the entry point and make its failure a
value rather than an exception. index.ts is now scaffolding plus
dispatch; view.ts owns the behaviour, returning Result<string, ViewError>
(a plain-data tagged union) via neverthrow, with try/catch confined to a
readFile adapter. The entry point matches the Result at the boundary.
Because viewFile is a pure in-process function it is now unit-tested
directly (and counted by coverage), while the subprocess tests stay as
the end-to-end check. ADR 0008 records the errors-as-values convention.
Drop the scaffolded test.todo placeholders (and the describe blocks left
empty by their removal), keeping only the six real parser tests. The
behaviours they enumerated live in the core-parser spec and tasks
0004-0007, which write their tests fresh when implemented.
Forward any positional arguments to `bun test` as path filters, so
`bun run test:report packages/core` reports just core while a bare
`bun run test:report` still covers the whole repo. The report's header
records the scope.
Add scripts/test-report.ts (bun run test:report): it runs the full suite
with JUnit + lcov output, then renders a gitignored reports/test-report.md
— a summary, a describe-grouped inventory (pass/todo/fail), and a
coverage table. The report is written even when tests fail and the script
exits with the test run's own status, so CI can gate on it later. Adds
fast-xml-parser (dev) for robust JUnit parsing; lcov is parsed by hand.
Add a bunfig.toml enabling coverage with text + lcov reporters, a
test:coverage script, and gitignore the coverage/ output. Test files and
the test-support helper are excluded so the numbers reflect product code
only.
packages/core/src/index.ts is a pure re-export barrel with no behavior
of its own, so its behavior tests belonged next to the module they
exercise. Relocate the parser tests (and their roadmap todos) into a
co-located parse_test.ts importing parse from ./parse.ts, and drop
index_test.ts. A broken re-export is still caught by typecheck and by
the bin package importing parse through the public entrypoint.
Drop the Point/Position types and the source/position fields from
Diagnostic. Nothing constructs a diagnostic yet — diagnostics is always
empty until task 0007 emits the first invalid-frontmatter warning, which
adds the locator fields back additively alongside its own tests.
Pull a reduced end-to-end smoke forward into the slice so the view flow
is exercised over a real on-disk recipe, not just a temp file.
Add fixtures/prose.md — a recipe using only this slice's modelled
constructs (frontmatter, headings, plain paragraphs) so it renders
losslessly. integration_test now runs `kitchen view fixtures/prose.md`
as a subprocess and asserts the frontmatter, separator, headings,
paragraphs, and flat document order end-to-end. smoke_test covers
`kitchen --help` exiting 0 with the view command listed.
The full fixtures/basic.md smoke remains task 0007, which needs 0004's
richer nodes before that fixture renders losslessly.
Cut the first complete thread through both layers with the smallest set
of node types.
@kitchen-md/core gains a pure, total `parse` returning a
`DocumentAST` of `{ frontmatter, blocks, diagnostics }`, built from a
minimal remark pipeline (remark-parse + remark-frontmatter) with an
internal translation layer to core's own types. This slice models
frontmatter passthrough, HeadingBlock, ParagraphBlock, and TextNode;
remark's mdast does not surface in the public API.
@kitchen-md/bin gains the `view` subcommand (commander) that reads a
file, calls `parse`, and passes the AST to a pure `render` that returns
an ANSI-styled string via chalk (auto-suppressed off a TTY). Frontmatter
prints as raw YAML followed by a separator, headings styled distinctly
by level, paragraphs as prose. A missing argument prints usage and a
missing/unreadable file a human-readable error, both exiting 1.
Richer blocks/inline plus raw fallbacks are task 0004; malformed
frontmatter diagnostics and the basic.md smoke test are task 0007.
