reqmd Test Specifications

The project shall build cleanly with go build ./... ./reqmd-code/... ./reqmd-import/... and all tests shall pass with go test ./... ./reqmd-code/... ./reqmd-import/... after every change.

Unit tests shall cover the parser, schema, and graph packages with inline fixtures (no external files): attr scanning and scalar typing, every graph check condition (broken, duplicate, ambiguous, circular, id-prefix, disposition, coverage, suppression), schema compilation paths (built-in injection, reserved-name rejection, additional-status-values validation), and the parent-child edge pass including broken-parent errors.

Unit tests SHALL verify frontmatter extraction into Document.Meta with first-file-wins merge semantics, goldmark body rendering for inline code, emoji, math, and fenced blocks, search-and-filter client logic (debounce, AND-combined filters, card hide/show, count update, child card visibility), and the server-rendered outline: correct parent/child nesting, expand/collapse, and scroll-spy targets.

Unit tests shall cover the content tree: preamble prose captured as a headingless info node, requirement-less files fully preserved, container headings promoted to KindContainer only when they have children, mid-file headings becoming nodes instead of body text, ParentID pointing at the nearest ancestor requirement (never a container), thematic breaks kept in bodies, a level-1 heading with an attr block rejected as a parse error, deterministic multi-file merge order, and the chunked parallel path retaining the file preamble. Tests shall also cover item-aware output: ls/stats Type columns and counts, CSV rows in document order for all node kinds, HTML container sections and info blocks with stable anchors, pageModel entries carrying type/parentId/children, and --filter pruning requirements while keeping items.

Integration tests shall run the full pipeline end-to-end: parse a multi-level requirement tree (including sub-requirements and items), validate all attributes, build the graph (including the parent-child pass), run the complete graph check catalog, and produce correct CSV/HTML output with hierarchical display. The repository's own spec/ tree shall be the primary fixture: it must validate with zero invalid requirements and must exercise level-typed coverage, document-level coverage defaults, verification results, and imported proxy documents.

Unit tests shall cover the four NDJSON tables: one row per content node in document order, built-in promotion to top-level columns, the attrs envelope for remaining attributes, null-valued fixed columns present on every row, trace rows carrying ref/target/pin/resolution state, results rows including evidence cases and omitting the table when no results are loaded, findings rows carrying level/code/direction, and byte-identical output across repeated exports of the same tree.

Tests shall cover the HTML export: server-rendered Document view with requirement cards, nested child cards, container sections and info blocks; attribute grid including undeclared attributes and array/nested values; coverage, verdict, and link-state chips including no verdict as a first-class outcome; escaped requirement text, attribute values, and embedded pageModel JSON; the project home page (catalog, health, actionable queues); and the analytical views. A browser-driven smoke suite shall verify deep links, URL filter round-tripping (including explicitly empty filters), outline selection, palette navigation, matrix bounds, and mobile outline behaviour.

Tests shall cover command wiring through cobra: check, ls, stats, export csv|html|json, serve, and baseline diff flag parsing; filter compile errors surfacing as command failures; --relaxed-filter warnings; --disjoint-check from both flag and schema; --ignore-unbound-results; --relaxed-versions demoting outdated pins while leaving predated pins fatal; and exit codes 0/1/2 for valid, invalid, and unparsable trees.

Unit tests shall cover init: the generic preset rendering schema and example with the requested id-prefix and title, the results preset defaulting to verify-results level and VR prefix, a custom preset directory rendering both Go templates with .ID, .Title, .Level, and .IDPrefix, and the run aborting on an existing document unless --force is given.

Unit tests shall cover version-pin delta computation and application: outdated refs proposed and rewritten in place, predated refs reported but never rewritten, --promote-unpinned adding unpinned refs only against a versioned upstream, whole-ref matching so a promote cannot rewrite the bare ID inside an already-pinned sibling, JSON delta shape and counts, and rewrite scope limited to ```attr blocks so surrounding prose is preserved byte-for-byte.

Unit tests shall cover the internal/filter package: expression compilation (including rejection of undeclared attributes), matching against requirement attribute maps plus id/title, per-document nil handling, FilterDocs/MatchingIDs, filter-aware coverage (a filtered-out inbound cannot satisfy a filtered-in requirement), and the --relaxed-filter path (BuildFoundAttrs, CompileRelaxed warnings).

Unit tests shall cover the disjoint-attribute graph check: zero intersection produces an ERROR, overlapping values pass, empty/absent attribute is exempt, enablement via --disjoint-check and x-reqmd.disjoint-check, and suppression via reqmd-suppress.

Unit tests shall cover the status lifecycle: missing status defaults to approved, draft requirements do not satisfy coverage, x-reqmd.additional-status-values validation (lowercase-only, reserved collision, duplicates), x-reqmd.ignore-status making all requirements coverage providers, and the draft-downstreams-ignored coverage message.

Unit tests shall cover level-typed requires-trace-from: a level token expands to every document declaring it, coverage is satisfied by any approved inbound from those documents, document-id tokens still resolve, and unknown tokens produce a WARNING.

Unit tests shall also cover the x-reqmd.requires-trace-from document default (SYS-FMT-003): a requirement that does not declare the attribute inherits the document default and is checked against it; a requirement that declares its own attribute overrides the default; requires-trace-from: [] opts a requirement out even under a non-empty default; an empty document default opts every requirement in the document out; and a document without a default leaves its requirements on the generic boundary-inference fallback.

Unit tests shall cover the internal/verify package: CTRF JSON parsing (status→outcome mapping, flaky→inconclusive, uninstrumented tests skipped silently, ~N pin preservation, malformed JSON and non-CTRF shape handling), multi-file latest-wins merge by timestamp, manual-results loading via parser.Discover, non-CTRF JSON skip, and pseudo-requirement synthesis (ID, trace with pin, outcome attr, synthetic document flag).

Unit tests shall cover the missing-verdict and failing-verdict graph checks: no-op when no result nodes are present, missing-verdict fires for approved measures without results, missing-verdict skips draft measures, failing-verdict fires for a fail outcome at ERROR, failing-verdict does not fire for pass, and version-pin fires on stale result pins.

Unit tests shall also cover case-keyed binding and evidence roll-up: extra.x-reqmd.case/verifies/description parsing and silent skip of uninstrumented tests, latest-wins per (measure, case) by stop, synthesized TC:<case> nodes with trace edges and description bodies, cased RESULT: identities, roll-up over authored and synthesized downstream test cases (mixed evidence with one failing case fails the requirement), draft downstream cases reported as ignored by missing-verdict, filter-aware evidence, deferred/rejected measure exemption, unbound-result warnings and their --ignore-unbound-results suppression, and the INFO verdict line plus the JSON/text results section.

Integration tests shall verify that serve --results <path> loads verification results, renders verdict badges in the HTML output, watches result file paths for changes (both fsnotify and polling fallback), and triggers a rebuild when result files change. Tests shall confirm that CTRF JSON changes update verdict badges from pass to fail and that the failing-verdict check fires on rebuild.

Unit tests shall cover the reqmd-code pipeline: language plugin parse/classification for C, C++, Go, Python, and Rust (kinds, method receivers, doc binding, top-level-only scope rules), trace extraction (explicit reqmd:trace markers and opt-in heuristic IDs), and the emitted exchange bundle (per-package documents with distinct id-prefixes, native-id-keyed requirements, method parents, provenance attrs, deterministic output across runs, and duplicate-symbol dedup). The bundle shall be decoded and rendered back through the shared exchange package so the producer/renderer contract is exercised end to end.

Unit tests shall cover the renderer: decoding (unknown field and trailing-data rejection), bundle validation (format mismatch, missing native-id or id-prefix, duplicate native-id, unknown parent or native-trace, reserved attr, escaping directory, parent cycle, duplicate output path), topology (children moved after their parents, multi-level heading depth), rendering (attr quoting and round-trip through a YAML parser, sorted deterministic attr order, unsupported nested attr rejected), schema generation (properties/required plus x-reqmd metadata), and idempotent writing (byte-identical re-runs, hand-authored files never clobbered).

Unit tests shall cover the reqmd_io package: snake_case fields serialized to the format's hyphenated keys, empty values omitted, numeric and boolean attribute values preserved, external: true preserved, and bundle JSON carrying the format version and writing to a file.