* test(#3696): failing-first coverage for the last_activity invariant and --strict exit status * fix(#3696): report the last_activity invariant and make the verdict gateable with --strict * fix(#3696): agree with the real reader on last_activity, and stop reporting structure as truncation * chore(#3696): backfill changeset PR number --------- Co-authored-by: sim <sim@local>
This commit is contained in:
Tom Boucher
@@ -214,6 +214,181 @@ function locateFieldRow(content: string, fieldName: string): { valueStart: numbe
|
||||
return null;
|
||||
}
|
||||
|
||||
/**
|
||||
* True only when y/m/d name a date that actually exists on the calendar.
|
||||
*
|
||||
* `Date.parse` validates shape but not value: it rolls an out-of-range day
|
||||
* FORWARD rather than rejecting it (`2026-02-30` -> `2026-03-02`,
|
||||
* `2026-04-31` -> `2026-05-01`). Shape-only validation would therefore
|
||||
* propagate a different, wrong instant instead of failing safe — precisely
|
||||
* what ADR-227 ("validate shape AND value; on failure of either layer coerce
|
||||
* to the contract's safe default, never propagate") exists to prevent. A
|
||||
* round-trip through Date.UTC detects the rollover: any component the
|
||||
* constructor normalised comes back changed.
|
||||
*
|
||||
* #3696: this predicate previously lived privately inside `smart-entry.cts`,
|
||||
* where it gated `parseActivityTimestamp`. `state validate` needed the same
|
||||
* answer to assert the `last_activity` invariant (S008), and a second copy is
|
||||
* the "generative fix divergence" class outright — two surfaces that disagree
|
||||
* about whether a STATE.md is usable is the defect #3696 opens with, so a
|
||||
* parity test over two copies would be codifying the bug rather than fixing
|
||||
* it. It moves here because this module is already the designated owner of
|
||||
* STATE.md field semantics (ADR-3180 §7.7) and `smart-entry.cts` imports no
|
||||
* peer that would make the reverse direction a cycle.
|
||||
*/
|
||||
/**
|
||||
* True when a field carries no value a writer ever supplied: absent, blank, or
|
||||
* still holding the shipped template's bracket placeholder.
|
||||
*
|
||||
* `templates/state.md:35` ships `Last activity: [YYYY-MM-DD] — [What happened]`,
|
||||
* so EVERY freshly-initialized project has this exact string until something
|
||||
* records activity. #3696's first cut only spared the ABSENT form, which made
|
||||
* S008 fire on the shipped template itself — caught by the pre-existing
|
||||
* "template-equivalent phase identities remain clean without disk drift" test,
|
||||
* which is precisely what it is there for.
|
||||
*
|
||||
* The placeholder test is anchored at the START rather than "contains a bracket
|
||||
* anywhere", so a real description that happens to cite one — `2026-08-19 — fixed
|
||||
* [#123] parsing` — is still a filled-in value. That keeps the rule from
|
||||
* silently swallowing genuine drift.
|
||||
*
|
||||
* Distinct from `isStateTemplateDefault`, which answers a different question
|
||||
* ("may a later handler overwrite this?") and deliberately returns true for a
|
||||
* bare ISO date — a perfectly valid value here.
|
||||
*/
|
||||
export function isUnfilledFieldValue(value: string | null | undefined): boolean {
|
||||
if (value === null || value === undefined) return true;
|
||||
const trimmed = value.trim();
|
||||
return trimmed === '' || trimmed.startsWith('[');
|
||||
}
|
||||
|
||||
/**
|
||||
* The `YYYY-MM-DD` prefix of `value`, but only when it names a date that
|
||||
* actually exists. `null` for anything else — no leading date token at all, or
|
||||
* a token that is shape-valid and calendar-impossible.
|
||||
*
|
||||
* #3696 review: this is deliberately a LEADING-TOKEN test, not the fully
|
||||
* anchored prose grammar `parseProseLastActivityField` uses. That function
|
||||
* requires the whole value to be `date` or `date <separator> description`, and
|
||||
* returns `{date: <the entire raw string>}` when it does not match — a shape
|
||||
* that reads like success. Asserting the S008 invariant through it therefore
|
||||
* rejected values the real reader accepts: `smart-entry`'s
|
||||
* `parseActivityTimestamp` needs only a leading date and reconstructs the
|
||||
* instant even when the suffix carries no dash, so
|
||||
* `Last activity: 2026-08-24 Shipped feature X` parses fine there while S008
|
||||
* called it unreadable. That is the same two-surfaces-disagree defect #3696
|
||||
* exists to close, merely pointing the other way.
|
||||
*
|
||||
* So the invariant asserted is the one the readers actually share: a leading
|
||||
* ISO date token that is a real calendar date.
|
||||
*/
|
||||
export function leadingCalendarDate(value: string | null): string | null {
|
||||
if (!value) return null;
|
||||
const match = /^(\d{4})-(\d{2})-(\d{2})(?![\d-])/.exec(value.trim());
|
||||
if (!match) return null;
|
||||
return isRealCalendarDate(Number(match[1]), Number(match[2]), Number(match[3]))
|
||||
? `${match[1]}-${match[2]}-${match[3]}`
|
||||
: null;
|
||||
}
|
||||
|
||||
export function isRealCalendarDate(year: number, month: number, day: number): boolean {
|
||||
if (month < 1 || month > 12 || day < 1 || day > 31) return false;
|
||||
const probe = new Date(Date.UTC(year, month - 1, day));
|
||||
return (
|
||||
probe.getUTCFullYear() === year &&
|
||||
probe.getUTCMonth() === month - 1 &&
|
||||
probe.getUTCDate() === day
|
||||
);
|
||||
}
|
||||
|
||||
/**
|
||||
* Markdown structure that can legitimately follow a single-line field. A line
|
||||
* matching any of these is the NEXT construct, never a continuation of the
|
||||
* field above it.
|
||||
*
|
||||
* BREADTH IS THE POINT, and the failure direction is deliberate: a missed
|
||||
* truncation costs a diagnostic nobody sees, while a false S009 reports drift on
|
||||
* a well-formed STATE.md — a gate that fires on valid documents is worse than no
|
||||
* gate. When a shape is ambiguous, it belongs here.
|
||||
*
|
||||
* #3696 review round 2 added the last three arms after all three were shown to
|
||||
* produce false S009 fires on well-formed content: an indented code block, an
|
||||
* HTML block, and a setext underline (`===`, which the `[-*_]{3,}` rule does not
|
||||
* cover — it only knows `-`, `*` and `_`).
|
||||
*/
|
||||
const MD_STRUCTURE_LINE_RE =
|
||||
/^(?:#{1,6}\s|\||>|```|~~~|[-*_]{3,}\s*$|=+\s*$|[-*+]\s|\d+[.)]\s|\[[^\]]+\]:|<|(?: {4}|\t))/;
|
||||
|
||||
/**
|
||||
* A setext heading's underline — `===` or `---` on its own line. The line ABOVE
|
||||
* one of these is a heading TITLE, which is indistinguishable from prose on its
|
||||
* own, so the scan must look ahead by one line rather than consume it. Without
|
||||
* this, `Last activity: …\nMy Heading\n===` reported "My Heading ===" as dropped
|
||||
* continuation text (#3696 review round 2).
|
||||
*/
|
||||
const SETEXT_UNDERLINE_RE = /^(?:=+|-+)\s*$/;
|
||||
|
||||
const STATE_SIBLING_FIELD_LINE_RE = /^\*{0,2}[A-Za-z][A-Za-z0-9 _-]*\*{0,2}:{1,2}\*{0,2}(?:\s|$)/;
|
||||
|
||||
/**
|
||||
* Return the prose that FOLLOWS a single-line field but plainly belongs to it —
|
||||
* i.e. the remainder `stateExtractField` silently drops when a writer emits a
|
||||
* value long enough to wrap.
|
||||
*
|
||||
* `stateExtractField`'s `(.+)` is newline-excluding, so
|
||||
*
|
||||
* Last activity: 2026-08-19 — Project initialized from ingest; PROJECT.md,
|
||||
* REQUIREMENTS.md, ROADMAP.md written
|
||||
*
|
||||
* yields only the first line and the rest is lost with no diagnostic (#3696).
|
||||
* `templates/state.md` prescribes a single-line field, so the DOCUMENT is what
|
||||
* is wrong here, not the reader — this function exists so `state validate` can
|
||||
* SAY so, not so the reader can start guessing at a multi-line grammar the
|
||||
* template does not sanction.
|
||||
*
|
||||
* That is also why the fix is not in `stateExtractField` itself: it has 20
|
||||
* direct callers and a CRITICAL blast radius (ADR-3180 §7.7, Rejected #1), and
|
||||
* joining continuations there would apply to every field — `Status:` would
|
||||
* swallow the line beneath it.
|
||||
*
|
||||
* Returns `null` when the field is absent, is a pipe-table row (a table cell
|
||||
* cannot wrap), or is followed by end-of-file, a blank line, Markdown
|
||||
* structure, or a sibling field.
|
||||
*/
|
||||
export function stateFieldContinuation(content: string, fieldName: string): string | null {
|
||||
const escaped = escapeRegex(fieldName);
|
||||
// Same two single-line grammars stateExtractField uses, in the same order, so
|
||||
// this locates exactly the line whose value it returned. The pipe-table rung
|
||||
// is deliberately absent: a `| Field | value |` row is bounded by its closing
|
||||
// pipe and cannot wrap.
|
||||
const match =
|
||||
new RegExp(`\\*\\*${escaped}:\\*\\*[ \\t]*(.+)`, 'i').exec(content) ??
|
||||
new RegExp(`^${escaped}:[ \\t]*(.+)`, 'im').exec(content);
|
||||
if (!match) return null;
|
||||
|
||||
// `(.+)` stops at the line terminator, so the field's line ends where the
|
||||
// match does. JS `.` excludes \r as well as \n, so on a CRLF document the \r
|
||||
// sits just AFTER the match rather than inside it — hence the strip below
|
||||
// before testing for the newline.
|
||||
const afterValue = match.index + match[0].length;
|
||||
const rest = content.slice(afterValue).replace(/^\r/, '');
|
||||
if (!rest.startsWith('\n')) return null; // end of file: nothing follows
|
||||
|
||||
const lines = rest.slice(1).split('\n').map((line) => line.replace(/\r$/, ''));
|
||||
const continuation: string[] = [];
|
||||
for (let i = 0; i < lines.length; i++) {
|
||||
const line = lines[i];
|
||||
if (!line.trim()) break;
|
||||
if (MD_STRUCTURE_LINE_RE.test(line)) break;
|
||||
if (STATE_SIBLING_FIELD_LINE_RE.test(line)) break;
|
||||
// Look ahead one line: a setext underline below makes THIS line a heading
|
||||
// title, so stop before consuming it rather than after.
|
||||
if (i + 1 < lines.length && SETEXT_UNDERLINE_RE.test(lines[i + 1])) break;
|
||||
continuation.push(line.trim());
|
||||
}
|
||||
return continuation.length ? continuation.join(' ') : null;
|
||||
}
|
||||
|
||||
export function stateExtractField(content: string, fieldName: string): string | null {
|
||||
const escaped = escapeRegex(fieldName);
|
||||
// Bold inline format: **FieldName:** value
|
||||
|
||||
Reference in New Issue
Block a user