@voxgig/sdkgen 4.21.0 → 4.23.0

package/dist/action/doctor.js

@@ -6,57 +6,17 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.action_doctor = action_doctor;
 exports.doctor = doctor;
 const kindCollection_1 = require("../helpers/kindCollection");
-// `voxgig-sdkgen doctor` — does this project's `.sdk/` still match the
-// scaffold?
-//
-// WHY THIS EXISTS
-//
-// Nothing told a project that its `.sdk/` had drifted. `target add`
-// OVERWRITES the vendored components, the template masters AND the per-target
-// model file, so any hand-edit there is silently reverted on the next run —
-// and any file the scaffold has since stopped shipping just stays, compiling
-// into the output forever. In voxgig-solardemo-sdk two such orphans broke the
-// Go build in a single session: `utility/make_target.go` (replaced upstream by
-// `make_point.go`, leaving a duplicate symbol) and `utility/struct/go.mod` (a
-// nested module that made `utility/struct` unimportable).
-//
-// Every divergence in that repo was found by hand-diffing. Without a check,
-// fixing it once guarantees nothing about next month.
-//
-// WHY A NAIVE `diff -r` DOES NOT WORK
-//
-// `target add` writes template masters with substitution PARTLY applied, and
-// inconsistently: `tm/go/core/error.go` arrives substituted (`SolardemoError`
-// where the scaffold says `ProjectNameError`), `tm/ts/test/utility.ts` arrives
-// raw (its placeholders are substituted later, at generate time), and
-// `tm/go/LICENSE` has the year filled in. A plain `diff -r` against the
-// scaffold reported 20 edited files in that repo; 19 were substitution
-// artefacts and exactly ONE was a real hand-edit.
-//
-// Only sdkgen knows which replacements it applied to which files, which is
-// why this check belongs here and cannot be scripted downstream. It re-runs
-// the same substitution before comparing, so what it reports is real.
 const node_path_1 = __importDefault(require("node:path"));
 const jostraca_1 = require("jostraca");
 const types_1 = require("../types");
 const utility_1 = require("../utility");
 const stdrep_1 = require("../helpers/stdrep");
 const target_1 = require("./target");
-// `aliasModelText` is no longer named here: the model-file comparison asks
-// the registry for the kind's own `rename`, so a kind that renames
-// differently is handled without this file knowing how.
 const kind_1 = require("./kind");
 const resolve_1 = require("./resolve");
 const definition_1 = require("../helpers/definition");
 const featureSource_1 = require("../helpers/featureSource");
 const junk_1 = require("../helpers/junk");
-// jostraca's Copy walk skips these (IGNORED_RE in CopyOp) — editor backups and
-// deliberately-disabled templates never reach a project, so they are not drift.
-//
-// `isJunk` is the other half of the same statement: what the WRITER refuses to
-// copy, this reader must refuse to miss. A `__pycache__` in the scaffold would
-// otherwise be reported as missing from every project on earth, and one in a
-// project as an unexplained extra file. See helpers/junk.
 const IGNORED_RE = /(~|-jostraca-off)$/;
 function ignoredEntry(name) {
     return IGNORED_RE.test(name) || (0, junk_1.isJunk)(name);
@@ -64,16 +24,6 @@ function ignoredEntry(name) {
 // Extensions jostraca copies byte-for-byte. Comparing them as text would
 // report spurious differences, so they are compared by raw bytes.
 const BINARY_RE = /\.(png|jpg|jpeg|gif|ico|pdf|zip|gz|woff2?|ttf|eot|wasm)$/i;
-// Root-level components sdkgen provides, and what a project loses by not
-// wiring one in. A project's `Root.ts` / `Top.ts` come from create-sdkgen at
-// init and are then FROZEN — `target add` never touches them — so a
-// capability added to sdkgen afterwards is invisible to every existing
-// project. solardemo's `Top.ts`, written before `ReadmeTop` existed, kept
-// hand-rolling a 9-line stub root README with an empty mermaid diagram for
-// months, and no number of `target add` runs would ever have said so.
-//
-// Not wiring one in is a legitimate choice, so this is REPORTED, never a
-// failure.
 const ROOT_COMPONENTS = [
     ['ReadmeTop', 'the assembled root README (quickstart, howto, test, package table)'],
     ['AgentGuideTop', 'the root AGENTS.md / CLAUDE.md agent guides'],
@@ -99,10 +49,6 @@ async function action_doctor(args, actx) {
 async function cmd_doctor_check(_args, actx) {
     return doctor(actx);
 }
-// `doctor prune` — the ONE write doctor performs, and only ever of files a
-// target's model DECLARES superseded (retired templates jostraca cannot
-// delete from a consumer: the old runner beside its replacement, a
-// pre-reshape vendored file). Everything else doctor does stays read-only.
 async function cmd_doctor_prune(_args, actx) {
     const log = actx.log;
     const fs = actx.fs();
@@ -120,10 +66,6 @@ async function cmd_doctor_prune(_args, actx) {
     });
     return { report: { pruned, ok: true } };
 }
-// Generated files the model says this toolchain NO LONGER writes, that
-// still exist in the project's target output directories. Paths in the
-// model are relative to each target's output dir (a sibling of `.sdk/`,
-// named for the target).
 function supersededFiles(actx) {
     const fs = actx.fs();
     const model = actx.model;
@@ -140,8 +82,49 @@ function supersededFiles(actx) {
     }
     return found;
 }
-// Code API. Returns the report; the CLI turns a non-ok report into a
-// non-zero exit so this can gate CI.
+function orphanModelFiles(actx) {
+    const fs = actx.fs();
+    const modeldir = node_path_1.default.join(actx.folder, 'model');
+    if (!fs.existsSync(modeldir)) {
+        return [];
+    }
+    const all = walk(fs, modeldir)
+        .filter((rel) => rel.endsWith('.aon') || rel.endsWith('.aontu'))
+        .filter((rel) => !rel.includes('.jostraca/') && !rel.startsWith('guide/'));
+    const ENTRY = [
+        'sdk.aon', 'sdk.aontu',
+        'test/test.aon', 'test/test.aontu',
+        '.model-config/model-config.aon', '.model-config/model-config.aontu',
+    ];
+    const seen = new Set();
+    const queue = ENTRY.filter((rel) => all.includes(rel));
+    while (0 < queue.length) {
+        const rel = queue.shift();
+        if (seen.has(rel)) {
+            continue;
+        }
+        seen.add(rel);
+        let src = '';
+        try {
+            src = fs.readFileSync(node_path_1.default.join(modeldir, rel), 'utf8');
+        }
+        catch (e) {
+            continue;
+        }
+        for (const m of src.matchAll(/@"([^"]+)"/g)) {
+            const ref = m[1];
+            if (ref.startsWith('@')) {
+                continue;
+            }
+            const from = node_path_1.default.posix.dirname(rel);
+            const next = node_path_1.default.posix.normalize(node_path_1.default.posix.join('.' === from ? '' : from, ref.replace(/^\.\//, '')));
+            if (all.includes(next) && !seen.has(next)) {
+                queue.push(next);
+            }
+        }
+    }
+    return all.filter((rel) => !seen.has(rel)).sort();
+}
 async function doctor(actx, scope) {
     const log = actx.log;
     const fs = actx.fs();
@@ -149,42 +132,17 @@ async function doctor(actx, scope) {
     const root = actx.folder;
     const report = {
         forked: [], edited: [], stale: [], missing: [], additive: [],
-        superseded: [], unwired: [], resyncPending: [], aliasedDiff: [], ok: true,
+        superseded: [], unwired: [], orphanModel: [],
+        resyncPending: [], aliasedDiff: [], ok: true,
     };
-    // Retired-output leftovers first: purely model-driven, and a finding —
-    // two copies of the same machinery, one stale, is exactly the state the
-    // migration guide had to clean up by hand three times.
     report.superseded = supersededFiles(actx);
-    // EVERY KIND, not just targets.
-    //
-    // `add` writes a copied model file for each kind — `model/target/<t>.aon`
-    // and `model/feature/<f>.aon` alike — and overwrites it on every resync.
-    // Only the target one was ever compared, so a hand-edit to an installed
-    // FEATURE definition read as perfectly in sync and was silently reverted by
-    // the next `target add` (which re-runs `feature add` for every active
-    // feature). That is exactly the failure `checkTargetModel` was added for,
-    // left open for the other half of what add owns.
-    //
-    // Driven off the registry, so a kind added later is checked without
-    // anything here changing. The per-kind EXTRAS — a target's `src/cmp` and
-    // `tm` trees — stay in that kind's own check, because they are genuinely
-    // different work rather than the same work with different strings.
+    report.orphanModel = orphanModelFiles(actx);
     const kinds = Object.keys(kind_1.KINDS).sort();
     const counts = {};
     for (const kind of kinds) {
         counts[kind] = Object.keys((0, kindCollection_1.kindCollection)(model, kind) ?? {}).length;
     }
     log.info({ point: 'doctor-start', targets: counts.target ?? 0, ...counts });
-    // Targets resolved ONCE, before the kind loop, because the feature checks
-    // need them too: a feature's per-target source lives under the name the
-    // target has in its own source, so finding it means knowing where each
-    // target came from. Resolving per (feature, target) pair would re-read
-    // every package manifest dozens of times and warn repeatedly about the
-    // same unresolvable target.
-    //
-    // Resolved even when the run is SCOPED past them — a scoped check of one
-    // feature still has to look in every target's tree for that feature's
-    // files.
     const targets = new Map();
     for (const tname of Object.keys(model?.main?.[types_1.KIT]?.target ?? {}).sort()) {
         const source = resolveDeclared('target', tname, actx);
@@ -218,15 +176,17 @@ async function doctor(actx, scope) {
             checkItemModel(actx, kind, source, report);
         }
     }
-    // Root-component wiring is a property of the PROJECT, not of any item, so a
-    // scoped run has no business reporting on it — `package update` asking "is
-    // this package's copy clean" should not also say the project never wired in
-    // `ReadmeTop`.
     if (null == scope) {
         checkWiring(actx, report);
     }
+    // orphanModel counts, and `unwired` does not. Not skipping a root
+    // component is a legitimate project choice that doctor only mentions; a
+    // model file nothing reads is never a choice - it is either a file that
+    // should be included, or one that should be deleted, and it reads as
+    // authoritative either way.
     report.ok = 0 === report.forked.length + report.edited.length +
-        report.stale.length + report.missing.length + report.superseded.length;
+        report.stale.length + report.missing.length + report.superseded.length +
+        report.orphanModel.length;
     for (const [kind, note] of [
         ['forked', 'FORKED (will be reverted by `target add`)'],
         ['edited', 'EDITED template master'],
@@ -235,6 +195,7 @@ async function doctor(actx, scope) {
         ['additive', 'additive (project-owned, not drift)'],
         ['superseded', 'SUPERSEDED generated output (run `doctor prune` to delete)'],
         ['unwired', 'NOT WIRED IN (root capability this project is missing)'],
+        ['orphanModel', 'ORPHAN MODEL FILE (on disk, included by nothing, read by nobody)'],
         ['resyncPending', 'RESYNC PENDING (predates provenance; `target add` updates it)'],
         ['aliasedDiff', 'aliased model differs from its origin (project-owned, not drift)'],
     ]) {
@@ -252,6 +213,7 @@ async function doctor(actx, scope) {
         additive: report.additive.length,
         superseded: report.superseded.length,
         unwired: report.unwired.length,
+        orphanModel: report.orphanModel.length,
         resyncPending: report.resyncPending.length,
         aliasedDiff: report.aliasedDiff.length,
         note: report.ok ?
@@ -287,30 +249,10 @@ function checkWiring(actx, report) {
         }
     }
 }
-// Where one declared item came from, or undefined with the reason said out
-// loud.
-//
-// The model records `base` and `origname` (written by add), and `origname` is
-// what makes an ALIAS checkable: the ref used to be rebuilt as
-// `<base>/../<name>`, which names the INSTALLED item — so `target add ts~ts2`
-// sent doctor looking for a `ts2` scaffold that does not exist, both trees
-// walked empty, and every component read as `additive` while every template
-// read as `stale` (a FAILING category). Real edits were undetectable and
-// doctor went red on noise.
-//
-// A bare name falls back to the bundled scaffold, for a copy predating
-// provenance.
 function resolveDeclared(kind, name, actx) {
     const declared = (0, kindCollection_1.kindCollection)(actx.model, kind)?.[name];
     const ref = (0, kind_1.recordedRef)(declared, name) || name;
     try {
-        // WITH THE LOG. Resolution reads the source's package manifest, and an
-        // unusable one is reported only through `ctx$.log`. Without it doctor
-        // silently lost the package name — and then reported every item from that
-        // package as FORKED, because the expected text no longer carries the
-        // `package:` line the copy has, with nothing anywhere saying why. That is
-        // accurate (the next add really would strip the line) but unactionable,
-        // which is the same defect as being wrong.
         return (0, resolve_1.resolveSource)(ref, kind, { folder: actx.folder, fs: actx.fs, log: actx.log });
     }
     catch (err) {
@@ -329,18 +271,6 @@ function checkTarget(actx, resolved, report) {
     const fs = actx.fs();
     const model = actx.model;
     const root = actx.folder;
-    // The two TREES `target add` owns, and the substitution it applies to each.
-    // (It owns one FILE as well — see checkTargetModel, below.)
-    //
-    //   src/cmp — copied verbatim, so a byte compare is the truth.
-    //   tm      — copied through jostraca's template(), with ProjectName.
-    // An ALIASED install renames every `<Cmp>_<origname>` component to
-    // `<Cmp>_<alias>` and rewrites the origin name inside the file (sibling
-    // imports, the __dirname-relative fragment path), because components are
-    // dispatched by the convention `cmp/<t>/Main_<t>`. doctor has to apply the
-    // same two transforms or it expects the ORIGIN names in the alias's folder
-    // and reports the entire tree as missing — 25 files for `go~go2`, none of
-    // them a real finding.
     const aliased = tname !== torigname;
     const renameCmp = aliased ?
         (rel) => (0, target_1.aliasCmpName)(rel, torigname, tname) : undefined;
@@ -378,22 +308,10 @@ function checkTarget(actx, resolved, report) {
     const excludes = (0, target_1.trimFeatures)({ log: quietLog(actx.log), fs: () => fs, folder: root, model }, tfolder, torigname, tname, features);
     compareTrees(actx, report, trees, {
         excludes,
-        // Only a TARGET has foreign feature source landing in its tree.
         foreign: (kind) => 'edited' === kind ?
             foreignFeatureSource(actx, resolved) : [],
     });
 }
-// COMPARE A SET OF TREES against the sources they were copied from.
-//
-// Extracted from `checkTarget` unchanged, because the docs kind needs exactly
-// this and a second copy of it is the defect this workstream keeps producing.
-// What differs per kind stays in the caller: which trees, what the copy
-// substituted, whether a trim excludes part of the source, and whether
-// anything foreign is expected to land in the tree.
-//
-// This is the mechanism behind the rule in CLAUDE.md — anything an add
-// writes, doctor must compare, or the next add silently reverts a project's
-// edit. A kind that declares trees and is not walked here would break it.
 function compareTrees(actx, report, trees, opts) {
     const fs = actx.fs();
     const model = actx.model;
@@ -406,30 +324,10 @@ function compareTrees(actx, report, trees, opts) {
         const scaffoldFiles = 'edited' === tree.kind ?
             walk(fs, tree.scaffold).filter((rel) => !excluded(rel, excludes)) :
             walk(fs, tree.scaffold);
-        // Where each file the project should have COMES FROM: the name it lands
-        // under, mapped to the absolute path of the file it was copied from.
-        //
-        // An absolute path rather than a tree-relative one, because not every
-        // expected file comes from THIS tree — see the foreign-feature union
-        // below.
         const landed = new Map(scaffoldFiles.map((rel) => [
             null == tree.rename ? rel : tree.rename(rel),
             node_path_1.default.join(tree.scaffold, rel),
         ]));
-        // FOREIGN FEATURE SOURCE.
-        //
-        // A feature supplied by a different package than the target ships its
-        // per-target source in ITS OWN `tm/<target>/` overlay, and `feature add`
-        // copies that into the project. Compared against the target's scaffold
-        // alone, those files are present in the project and absent upstream — so
-        // every one of them was reported STALE, a FAILING category. Any project
-        // using an external feature had a red `doctor` and nothing wrong with it.
-        //
-        // They are not stale, they are expected — just expected from somewhere
-        // else. Adding them here rather than suppressing them keeps them
-        // CHECKED: a hand-edit to a foreign feature's source is still reported,
-        // which is what `package update`'s gate needs in order to cover
-        // everything the re-add writes.
         const foreign = new Set();
         for (const [rel, from] of (opts?.foreign?.(tree.kind) ?? [])) {
             landed.set(rel, from);
@@ -475,17 +373,6 @@ function compareTrees(actx, report, trees, opts) {
         }
     }
 }
-// A DOCS item's trees.
-//
-// Simpler than a target's in every way that matters here: no trim (a docs
-// item has no per-feature source to leave out), and nothing foreign lands in
-// them (a feature package's overlay targets a TARGET's tree). What is the
-// same is the alias handling — docs components are dispatched by the same
-// `Main_<n>` convention — and the tree paths, which come from the registry
-// rather than being spelled a second time.
-//
-// The optional template tree is skipped when the SOURCE does not ship one:
-// `edition add` did not copy it, so the project is right not to have it.
 function checkEdition(actx, resolved, report) {
     const fs = actx.fs();
     const root = actx.folder;
@@ -513,20 +400,6 @@ function checkEdition(actx, resolved, report) {
     });
     compareTrees(actx, report, trees);
 }
-// Per-target source that a FEATURE package supplies for this target, as
-// `project-relative path -> the file it was copied from`.
-//
-// Only for features whose source folder is not the target's own: everything
-// the target ships for itself is already in the tree being walked, and
-// listing it twice would compare it against itself.
-//
-// Only for ACTIVE features, because that is what `feature add` copies. A
-// feature switched off later leaves its files behind, and those really are
-// stale — reporting them is the point.
-//
-// Discovered with `findFeatureSources`, the same function the fan-out and the
-// trim use, so doctor's idea of which files belong to a feature cannot drift
-// from the one that put them there.
 function foreignFeatureSource(actx, target) {
     const out = new Map();
     const features = actx.model?.main?.[types_1.KIT]?.feature ?? {};
@@ -544,18 +417,6 @@ function foreignFeatureSource(actx, target) {
     }
     return out;
 }
-// One feature's per-target source, as `tm-relative path -> the file it was
-// copied from`, for source the FEATURE package supplies rather than the
-// target.
-//
-// ONE enumerator, used from both directions: the target walk needs these to
-// know they are expected, and the feature check needs them to compare. Two
-// enumerations would be two ideas of which files a feature owns, and the
-// pair would drift — which is the failure this codebase keeps producing.
-//
-// Empty when the feature's source IS the target's own tree: those files are
-// already in the tree being walked, and listing them would compare them
-// against themselves.
 function overlayFiles(actx, feature, target) {
     const fs = actx.fs();
     const out = new Map();
@@ -581,18 +442,6 @@ function overlayFiles(actx, feature, target) {
     }
     return out;
 }
-// A FEATURE's per-target source, checked from the feature's side.
-//
-// This is where those files are COMPARED. The target walk only marks them
-// expected, so that each is compared exactly once — by the feature that owns
-// it, which is also the only scope in which `feature add` will rewrite it.
-//
-// Doing it the other way round — leaving the comparison in the target walk —
-// left the gate open in the case it was built for: `package update` on a
-// FEATURE package scopes doctor to that feature, no target is walked, and so
-// nothing compared the very overlay the feature package supplies. Verified
-// before fixing: the update succeeded without asking, and the local edit was
-// destroyed.
 function checkFeatureSource(actx, feature, targets, report) {
     const fs = actx.fs();
     const root = actx.folder;
@@ -613,51 +462,15 @@ function checkFeatureSource(actx, feature, targets, report) {
         }
     }
 }
-// The copied MODEL FILE — `model/<kind>/<name>.aon`, written by add with
-// the `'BASE'` replacement, and overwritten on every resync exactly as
-// `src/cmp` and `tm` are.
-//
-// It is not scaffolding trivia. A target's carries its dependency set, its
-// `phase` gates, `srcfeature`, `feature.trim` / `feature.fullset` and its
-// publish-registry identity; a feature's carries its version, `active`
-// default, `config` defaults and hook wiring. So a maintainer who fixes a dep
-// version there loses the fix on the next resync with nothing said (the SDK
-// regresses with nobody touching it, which is the failure the "a project
-// decision belongs in the MODEL" rule exists to prevent), and a project whose
-// copy predates a scaffold change carries the old declaration forever while
-// doctor reports it in sync.
-//
-// KIND-NEUTRAL, because the file is the same kind of thing for every kind:
-// one definition, copied, stamped with provenance, optionally renamed for an
-// alias. Only the alias rewrite is kind-specific, and the registry already
-// says which kinds can be aliased at all.
 function checkItemModel(actx, kind, source, report) {
     const name = source.name;
     const origname = source.origname;
     const base = source.base;
     const fs = actx.fs();
     const scaffold = (0, definition_1.definitionPath)(source.folder, kind, origname);
-    // Nothing to compare against — a source that no longer ships this item.
     if (!fs.existsSync(scaffold)) {
         return;
     }
-    // An ALIAS's model file is PROJECT-OWNED: `target add go~go2` creates it and
-    // never overwrites it again, because differentiating it is the whole point
-    // of an alias (a second Go module needs its own module name and deps), and
-    // add-a-target tells the project to edit it. So a difference here is not a
-    // fork — `target add` will not revert it — and must not fail the check.
-    //
-    // It is still worth REPORTING, which it never was before: the origin was
-    // unrecoverable, so doctor skipped the file entirely and an alias that had
-    // drifted far from a moved-on upstream said nothing. With `origname`
-    // recorded the comparison is possible, so it runs, with the same key
-    // rewrite the add applied — and anything left over is the project's own
-    // differentiation, reported as informational.
-    //
-    // Only for a kind that CAN be aliased. A feature cannot, so for features
-    // this is always false and the ownership argument never applies — the
-    // registry says which, rather than this inferring it from the names
-    // happening to match.
     const aliased = (0, kind_1.kindDef)(kind).alias && name !== origname;
     const project = (0, definition_1.definitionPath)(actx.folder, kind, name);
     const label = 'model/' + kind + '/' + name + '.aon';
@@ -665,22 +478,7 @@ function checkItemModel(actx, kind, source, report) {
         report.missing.push(label);
         return;
     }
-    // The substitution the add applied on the way in: jostraca's template()
-    // against the model (so `module: name: '$$name$$'` arrives as the project
-    // slug) plus the provenance block recording where the scaffold was.
-    //
-    // NOT templateReplacements() — that is the tm/ tree's map. The two writers
-    // pass different maps, so doctor has to as well, or every project reads as
-    // forked on the `base:` line alone. The map itself is shared with the
-    // writer (helpers/stdrep) so the two cannot drift.
-    //
-    // `package` comes from the SOURCE's manifest, not from what the copy
-    // records, for the same reason `base` does: what this compares is what the
-    // add would write NOW. A package that renamed itself therefore reads as
-    // forked, which is accurate — the next add would rewrite the line.
     const provenance = (0, stdrep_1.provenanceReplace)({ base, origname, name, package: source.package });
-    // For an alias, compare against what the origin WOULD produce under the new
-    // name, so the rename itself is not the difference.
     const rename = (0, kind_1.kindDef)(kind).rename;
     const rewrite = (aliased && null != rename) ?
         (src) => rename(src, origname, name) : undefined;
@@ -691,52 +489,12 @@ function checkItemModel(actx, kind, source, report) {
         report.aliasedDiff.push(label);
         return;
     }
-    // A copy written before the toolchain stamped some provenance key differs
-    // from the scaffold by exactly the lines carrying that key. That is not a
-    // fork — the project changed nothing — and reporting it as one would turn
-    // every existing consumer's CI red on upgrade.
-    //
-    // PER KEY, not all-or-nothing. The keys arrived in stages — `base` and
-    // `origname` first, `package` with the manifest — so a project that
-    // resynced between two of them holds a copy that IS stamped (it has `base`)
-    // and is still missing a later key. An all-or-nothing "is it stamped at
-    // all" test called every one of those a fork: on the released 3.4.8
-    // scaffold only ts, csharp and swift carried the anchor, and `ts` is in
-    // essentially every consumer SDK, so that was close to the whole installed
-    // base going red on a file nobody touched.
     if (stampOnly(fs, scaffold, project, actx.model, provenance, rewrite)) {
         report.resyncPending.push(label);
         return;
     }
     report.forked.push(label);
 }
-// Is the ONLY difference a provenance line the copy has not been given yet?
-//
-// MATCHED AGAINST THE EXACT LINES THE STAMP PRODUCES, never against a pattern
-// for "a line that looks like provenance". `base`, `origname` and `package`
-// are not reserved words: `main: kit: target: <t>: module: package` (the Go
-// root package identifier) and `publish: registry: package` (the published
-// package name) are declared model slots that a target model may write in
-// block form, on their own line, looking exactly like a provenance line to
-// any regex. A generic pattern got BOTH directions wrong on such a model —
-// it read the copy's `module: package:` as proof the copy was already
-// stamped, so an untouched pre-manifest file was reported as forked; and it
-// stripped that same real line from the comparison, so DELETING it — an
-// actual fork, silently reverted by the next `target add` — was reported as
-// a pending resync and passed the check.
-//
-// Comparing against `provenanceReplace`'s own output has no such ambiguity,
-// and keeps the reader tied to the writer, which is the whole point of
-// helpers/stdrep.
-//
-// The rule, once the two are known to differ:
-//
-//   - every line the EXPECTED has and the copy lacks must be a stamp line;
-//   - the copy must have NO line the expected lacks.
-//
-// The second half is what keeps the tolerance narrow, as the previous
-// rollout's review required: a changed `package:` value leaves the old line
-// in the copy and unmatched, so it is a fork and is reported as one.
 function stampOnly(fs, scaffoldPath, projectPath, model, provenance, rewrite) {
     const { expected, actual } = renderPair(fs, scaffoldPath, projectPath, model, provenance, rewrite);
     // The rendered block, as lines: `base: '...'` plus whichever of
@@ -750,7 +508,6 @@ function stampOnly(fs, scaffoldPath, projectPath, model, provenance, rewrite) {
         0 < onlyExpected.length &&
         onlyExpected.every((line) => stamp.has(line.trim()));
 }
-// Lines of `a` that `b` does not have, counting duplicates.
 function lineDiff(a, b) {
     const pool = new Map();
     for (const line of b.split('\n')) {
@@ -801,14 +558,6 @@ function excluded(rel, excludes) {
     }
     return false;
 }
-// Compare a scaffold file with a project file, applying the SAME substitution
-// `target add` applied on the way in. Without this the comparison reports
-// every substituted placeholder as an edit.
-// `ignore` drops matching lines from BOTH sides before comparing, for a
-// difference that is known not to be a fork (see the provenance rollout in
-// checkTargetModel). It is deliberately a second, narrower question asked
-// only after a plain comparison has already found a difference — so a file
-// that matches exactly never depends on it.
 function differs(fs, scaffoldPath, projectPath, model, replace, ignore, rewrite) {
     if (BINARY_RE.test(scaffoldPath)) {
         return !fs.readFileSync(scaffoldPath).equals(fs.readFileSync(projectPath));
@@ -820,20 +569,6 @@ function differs(fs, scaffoldPath, projectPath, model, replace, ignore, rewrite)
     const strip = (s) => s.split('\n').filter((line) => !ignore(line)).join('\n');
     return strip(expected) !== strip(actual);
 }
-// What `target add` WOULD write, beside what the project actually has.
-//
-// One definition, shared by the byte comparison and the stamp-only tolerance,
-// so the two can never disagree about what the expected text is.
-//
-// template() runs even when there is nothing to REPLACE, because jostraca's
-// Copy always interpolates `$$ref$$` against the model as well — the replace
-// map is an extra, not the whole substitution. Skipping it for the src/cmp
-// tree (whose Copy passes no replace map) meant the three Config fragments
-// that carry `$$const.Name$$` / `$$main.kit.info.servers.0.url$$` arrived
-// substituted and compared as bytes: every project with ts, js or dart
-// reported `src/cmp/<t>/fragment/Config.fragment.<ext>` as FORKED straight
-// out of `target add`, which is precisely the noise this exists to remove.
-// With no `$$` refs and no replace keys, template() is the identity.
 function renderPair(fs, scaffoldPath, projectPath, model, replace, rewrite) {
     const rawsrc = fs.readFileSync(scaffoldPath, 'utf8');
     const src = null == rewrite ? rawsrc : rewrite(rawsrc);
@@ -842,7 +577,6 @@ function renderPair(fs, scaffoldPath, projectPath, model, replace, rewrite) {
         actual: fs.readFileSync(projectPath, 'utf8'),
     };
 }
-// trimFeatures logs its decisions; doctor is a report, not a run.
 function quietLog(log) {
     const noop = () => { };
     const quiet = { info: noop, debug: noop, warn: log.warn.bind(log), error: noop, trace: noop, fatal: noop };