pi-lens 4.2.1 → 4.4.0

package/dist/clients/config-core/index.js

@@ -1,50 +0,0 @@
-/**
- * THE shared config core (#2425). Every future config loader, catalog, and
- * selector resolves through this module rather than growing a fourth merge
- * semantics of its own.
- *
- * The pipeline, in one line:
- *
- *   RawConfig -> validate(schema) -> NormalizedConfig -> merge(sources) -> Resolved<T>
- *
- * `resolveConfig` runs both halves and is the front door. It is a pure function
- * of its arguments: no file reads, no ledger writes, no logging. Reporting is a
- * separate, explicit step — `reportPiLensConfigRecords` in
- * `clients/config-resolve.ts` — so a caller decides when a user gets warned and
- * under which subsystem; a library that warns on its own would fragment the
- * warn-once latch across every consumer.
- *
- * That step used to live HERE, in `records.ts`, which meant this "pure" library
- * imported `config-warn.js` and the degradation ledger (#2426). Purity was the
- * claim and the import graph disagreed: it broke every suite that `vi.mock`s the
- * ledger wholesale, and it closed seven import cycles the moment the first
- * loader — all of which sit downstream of `file-utils.ts` — resolved through the
- * core. The reporting function moved to the module that owns the loaders'
- * subsystem vocabulary, and the one constant records needs from the ledger
- * (`DEGRADATION_ENTRIES_PER_KIND`) moved to the `ledger-bounds.js` leaf that
- * exists for exactly this. Nothing under `config-core/` imports a sink now.
- *
- * ALL THREE LOADERS ARE WIRED as of #2426, through `clients/config-resolve.ts`
- * and the canonical schema in `clients/config-schema.ts`:
- * `clients/lsp/config.ts` (`loadLSPConfig`), `clients/lens-config.ts`
- * (`loadPiLensGlobalConfig`), and `clients/project-lens-config.ts`
- * (`loadPiLensProjectConfig`) are now projections of one resolution. #2416
- * brings the `lsp.servers.<id>` shape the schema currently only reserves.
- */
-export { SOURCE_TIERS } from "./provenance.js";
-// `merge()` itself is NOT re-exported. Its input type only PROMISES a
-// post-`validate()` value; nothing in the language enforces that a caller
-// who imports it directly honors the promise, and `merge()`'s own bounds are
-// a narrow backstop, not a second validator (see `merge.ts`'s module doc).
-// `resolveConfig` (re-exported at the bottom of this file, from `resolve.js`)
-// is the one supported way in: it always validates
-// every source before merging. `merge()` stays exported from `merge.ts`
-// itself — marked `@internal` there — for this module's own use and for
-// tests that probe it directly.
-// `resolveConfig` and its types live in `./resolve.js` (#2426) so a caller can
-// import the front door without also importing this barrel's width — in
-// particular `process-spec.js` -> `project-trust.js`, which closed three
-// import cycles once the config loaders (all downstream of `file-utils.ts`)
-// started resolving through the core. The public surface is unchanged: this is
-// still where the supported entry point is exported from.
-export { resolveConfig } from "./resolve.js";

package/dist/clients/config-core/merge.js

@@ -1,357 +0,0 @@
-/**
- * Field-wise merge with per-leaf provenance (#2425, scope items 2 and 3).
- *
- * The three loaders this replaces each merge differently: `lsp/config.ts`
- * spreads two objects and hand-patches four keys, `lens-config.ts` projects
- * field by field, and `project-lens-config.ts` never merges with the global
- * config at all. One merger with a per-node strategy expresses all three, and
- * — unlike any of them — it can say afterwards where each value came from.
- *
- * Rules:
- *
- * - OBJECTS ARE ALWAYS FIELD-WISE. A nearer tier that sets one key of an object
- *   does not erase the rest. This is #2415's "never whole-object replacement".
- * - ARRAYS follow the node's `x-merge-strategy`: `replace` by default, or
- *   `append`, or `keyed:<field>`.
- * - DENY-ANNOTATED NODES go to `deny.ts` instead, whatever the merge strategy
- *   says. Monotonic deny outranks precedence by construction.
- * - EVERY LEAF GETS PROVENANCE, keyed by its JSON-pointer path in the RESOLVED
- *   value, so a consumer navigates from an entry to the value it describes.
- *
- * Sources are sorted by tier precedence rather than trusted in caller order, so
- * assembling them in a different order cannot silently invert precedence. Ties
- * keep caller order, which is how several `nested-project` files stay ordered
- * from outermost to innermost.
- *
- * `merge()`'S CONTRACT IS NARROWER THAN "IT ENFORCES THE TWO BOUNDS TOO", which
- * an earlier version of this comment claimed (#2440 review round 3, finding
- * F8). It required
- * every source's `value` to already be `validate()`'s output — a sentence on
- * `ConfigSource`, not a compiler check — and does NOT independently re-walk a
- * value to confirm that. Five arms hand a contributor's value back BY
- * REFERENCE, never walked: the leaf fall-through in `mergeNode`, the
- * no-object arm in `mergeObject`, the no-array arm in `mergeArray`, `replace`,
- * and `appendArrays`. Any of those can be reached with an opaque schema node
- * whose tiers disagree on shape (one contributes an array, another a deeply
- * nested object at the same key) — `isObjectNode`/`isArrayNode` both answer
- * "no" for a mixed-shape node, the recursion falls through to the leaf arm,
- * and whatever depth or hostile keys live inside that value ride along
- * unexamined and unrecorded. Closing those five arms would mean re-deriving
- * `normalize.ts`'s walk a second time here — the "half-enforcement" this
- * function used to attempt and get half right.
- *
- * The fix is the CONTRACT, not a second walk: `merge()` is exported from this
- * file (`@internal`) for `resolveConfig` and for tests that probe it
- * directly, but it is not part of `index.ts`'s public surface. `resolveConfig`
- * is the one supported way in, and it validates every source before merging —
- * a mutation test in `hostile-input.test.ts` proves that by skipping
- * `validate()` on one source and watching the suite go red.
- *
- * The depth counter (`mergeNode`) and the prototype-key check (`mergeObject`,
- * over the KEYS an already-validated object's field-wise merge is about to
- * combine) still run. They are a backstop for the common, well-typed
- * recursion — a value `validate()` already bounded and sanitized, or a
- * same-shape hand-built value in a direct `merge()` test — not an
- * independent enforcement boundary; the paragraph above is what they do NOT
- * cover. The key check also still matters for one real case even after
- * `validate()`: a hostile key one tier introduces at a node the OTHER tier
- * also contributes to is a key `mergeObject` discovers only when it unions
- * that node's field names across contributions, so it is the one place that
- * can still name and record it, even though `safeAssign` alone would already
- * have refused the write.
- *
- * Pure: no state, no I/O, no ledger writes. Records, when the caller supplies a
- * collector, describe what the two backstops refused.
- */
-import { denyMemberProvenance, denyProvenance, resolveArrayDeny, resolveBooleanDeny, } from "./deny.js";
-import { isUnsafeConfigKey, MAX_CONFIG_DEPTH, safeAssign, UNSAFE_KEY_REASON, } from "./safe-object.js";
-import { boundedKeyLabel, MigrationRecordCollector, migrationSubject, } from "./records.js";
-import { tierPrecedence, } from "./provenance.js";
-import { denyPolicyOf, isConfigObject, isKnownSchemaType, isPlainObject, isSchemaNode, itemsSchema, keyedField, mergeStrategyOf, propertySchema, schemaType, } from "./schema.js";
-/**
- * Merge normalized sources into one resolved config with per-leaf provenance.
- *
- * Sources whose value is `undefined` contribute nothing — that is how a config
- * file that failed validation drops out without a special case.
- *
- * @internal Requires every source's `value` to already be `validate()`'s
- * output. Not part of `clients/config-core/index.ts`'s public surface —
- * `resolveConfig` is. See this module's doc comment for what merging a
- * hand-built value can still get past.
- */
-export function merge(sources, schema, options = {}) {
-    const ordered = [...sources]
-        .map((source, order) => ({ source, order }))
-        .sort((left, right) => {
-        const byTier = tierPrecedence(left.source.tier) - tierPrecedence(right.source.tier);
-        return byTier !== 0 ? byTier : left.order - right.order;
-    })
-        .map((entry, rank) => ({ source: entry.source, rank }));
-    const contributions = [];
-    for (const entry of ordered) {
-        // A source whose value is `undefined` contributes nothing. Filtering with
-        // a narrowing loop rather than `.filter()` keeps the element type exact:
-        // `Array.filter` cannot drop `undefined` from the value's type on its own.
-        const value = entry.source.value;
-        if (value === undefined)
-            continue;
-        contributions.push({ source: entry.source, value, rank: entry.rank });
-    }
-    const provenance = new Map();
-    const context = {
-        provenance,
-        collector: options.collector ?? new MigrationRecordCollector(),
-        file: options.file ?? "",
-    };
-    const value = mergeNode(contributions, schema, "", context, 0);
-    return { value: value, provenance };
-}
-/** Record one node the merger's own bounds refused. Never carries a value. */
-function record(context, entry) {
-    const key = boundedKeyLabel(entry.key);
-    context.collector.add({
-        code: entry.code,
-        file: context.file,
-        key,
-        subject: migrationSubject(context.file, key),
-        reason: entry.reason,
-    });
-}
-function provenanceOf(contribution, key) {
-    const { source } = contribution;
-    return {
-        tier: source.tier,
-        key,
-        ...(source.file === undefined ? {} : { file: source.file }),
-        ...(source.trust === undefined ? {} : { trust: source.trust }),
-    };
-}
-function stamp(context, contribution, key) {
-    context.provenance.set(key, provenanceOf(contribution, key));
-}
-function mergeNode(contributions, schema, key, context, depth) {
-    if (contributions.length === 0)
-        return undefined;
-    // The merger's own depth bound. `validate()` already truncated anything it
-    // walked, so this fires only for a value that reached `merge()` without it —
-    // which the exported signature permits and a review probe demonstrated.
-    if (depth > MAX_CONFIG_DEPTH) {
-        record(context, {
-            code: "PILENS_CFG_0005",
-            key,
-            reason: `config nesting exceeds ${MAX_CONFIG_DEPTH} levels; ignored`,
-        });
-        return undefined;
-    }
-    const denyPolicy = denyPolicyOf(schema);
-    if (denyPolicy)
-        return mergeDeny(contributions, denyPolicy, key, context);
-    if (isObjectNode(schema, contributions)) {
-        return mergeObject(contributions, schema, key, context, depth);
-    }
-    if (isArrayNode(schema, contributions)) {
-        return mergeArray(contributions, schema, key, context, depth);
-    }
-    const winner = contributions[contributions.length - 1];
-    stamp(context, winner, key);
-    return winner.value;
-}
-/**
- * An object node: either the schema says so, or every contribution IS a plain
- * object. The second arm matters because a schema may be opaque about a
- * sub-object, and merging two objects whole would drop the field-wise rule the
- * moment a schema is incomplete.
- */
-function isObjectNode(schema, contributions) {
-    const declared = schemaType(schema);
-    if (declared === "object")
-        return true;
-    // Only a RECOGNIZED non-object type rules object-ness out. An unrecognized
-    // keyword leaves the node opaque, and an opaque node is decided by the
-    // value's shape — the same rule `normalize.ts` walks by, so one schema typo
-    // cannot give the two halves different merge semantics.
-    if (isKnownSchemaType(declared))
-        return false;
-    if (schema && isSchemaNode(schema.properties))
-        return true;
-    return contributions.every((entry) => isPlainObject(entry.value));
-}
-function isArrayNode(schema, contributions) {
-    const declared = schemaType(schema);
-    if (declared === "array")
-        return true;
-    if (isKnownSchemaType(declared))
-        return false;
-    return contributions.every((entry) => Array.isArray(entry.value));
-}
-function mergeDeny(contributions, policy, key, context) {
-    const denyContributions = contributions.map((entry) => ({
-        tier: entry.source.tier,
-        value: entry.value,
-        ...(entry.source.file === undefined ? {} : { file: entry.source.file }),
-        ...(entry.source.trust === undefined ? {} : { trust: entry.source.trust }),
-    }));
-    const resolution = policy === "boolean-false"
-        ? resolveBooleanDeny(denyContributions)
-        : resolveArrayDeny(denyContributions);
-    const entry = denyProvenance(denyContributions, resolution, key);
-    if (entry)
-        context.provenance.set(key, entry);
-    // Plus one entry per surviving MEMBER of a union (#2427 review round 2,
-    // F2). The array's own entry answers "which tier denied first"; a caller
-    // asking about one member — "why is THIS server off" — needs the tier that
-    // contributed that member, and reading the array entry for it reported the
-    // wrong tier for every member after the first.
-    for (const member of denyMemberProvenance(denyContributions, resolution, key)) {
-        context.provenance.set(member.key, member);
-    }
-    return resolution.value;
-}
-function mergeObject(contributions, schema, key, context, depth) {
-    const objects = contributions.filter((entry) => isConfigObject(entry.value));
-    if (objects.length === 0) {
-        const winner = contributions[contributions.length - 1];
-        stamp(context, winner, key);
-        return winner.value;
-    }
-    // Field order: lowest tier's keys first, each nearer tier appending only the
-    // keys it introduces. A resolved config therefore reads in the order the
-    // user's own files introduced things.
-    //
-    // A prototype-modifying name is refused HERE, before it can become a child
-    // key: dropping it at the assignment alone would still have walked it and
-    // stamped provenance at a pointer no resolved value can be read from.
-    const names = [];
-    for (const entry of objects) {
-        for (const name of Object.keys(entry.value)) {
-            if (names.includes(name))
-                continue;
-            if (isUnsafeConfigKey(name)) {
-                record(context, {
-                    code: "PILENS_CFG_0006",
-                    key: `${key}/${name}`,
-                    reason: UNSAFE_KEY_REASON,
-                });
-                continue;
-            }
-            names.push(name);
-        }
-    }
-    const out = {};
-    for (const name of names) {
-        const childKey = `${key}/${name}`;
-        const childContributions = [];
-        for (const entry of objects) {
-            const parent = entry.value;
-            if (!Object.hasOwn(parent, name))
-                continue;
-            childContributions.push({
-                source: entry.source,
-                value: parent[name],
-                rank: entry.rank,
-            });
-        }
-        const merged = mergeNode(childContributions, propertySchema(schema, name), childKey, context, depth + 1);
-        if (merged !== undefined)
-            safeAssign(out, name, merged);
-    }
-    return out;
-}
-function mergeArray(contributions, schema, key, context, depth) {
-    const arrays = contributions.filter((entry) => Array.isArray(entry.value));
-    if (arrays.length === 0) {
-        const winner = contributions[contributions.length - 1];
-        stamp(context, winner, key);
-        return winner.value;
-    }
-    const strategy = mergeStrategyOf(schema);
-    const keyed = keyedField(strategy);
-    if (keyed !== undefined) {
-        return mergeKeyedArray(arrays, schema, keyed, key, context, depth);
-    }
-    if (strategy === "append")
-        return appendArrays(arrays, key, context);
-    // `replace`: the highest-precedence contributor supplies the whole array,
-    // and one provenance entry at the array's own pointer says who. Per-element
-    // entries would be noise: every element has the same answer.
-    const winner = arrays[arrays.length - 1];
-    stamp(context, winner, key);
-    return [...winner.value];
-}
-function appendArrays(arrays, key, context) {
-    const out = [];
-    for (const entry of arrays) {
-        for (const member of entry.value) {
-            stamp(context, entry, `${key}/${out.length}`);
-            out.push(member);
-        }
-    }
-    // The array itself is attributed to its lowest-precedence contributor: with
-    // `append`, that is the tier the list STARTED at.
-    stamp(context, arrays[0], key);
-    return out;
-}
-/**
- * `keyed:<field>` — match entries across tiers by `<field>` and merge them
- * field-wise, so a project file can override one server's `command` without
- * restating the whole catalog entry, and without erasing the sibling entries.
- *
- * Entries missing the key field cannot be matched, so they are appended in
- * contribution order and attributed to their own source. Dropping them would be
- * silent data loss; matching them together would merge unrelated entries.
- */
-function mergeKeyedArray(arrays, schema, field, key, context, depth) {
-    const items = itemsSchema(schema);
-    const order = [];
-    const groups = new Map();
-    const unkeyed = [];
-    for (const entry of arrays) {
-        for (const member of entry.value) {
-            const identity = keyIdentity(member, field);
-            const memberContribution = {
-                source: entry.source,
-                value: member,
-                rank: entry.rank,
-            };
-            if (identity === undefined) {
-                unkeyed.push(memberContribution);
-                continue;
-            }
-            const group = groups.get(identity);
-            if (group) {
-                group.push(memberContribution);
-                continue;
-            }
-            order.push(identity);
-            groups.set(identity, [memberContribution]);
-        }
-    }
-    const out = [];
-    const pushMerged = (group) => {
-        const merged = mergeNode(group, items, `${key}/${out.length}`, context, depth + 1);
-        // An entry can only merge to `undefined` when its group is empty, which
-        // the construction above cannot produce; dropping it keeps the resolved
-        // list free of holes either way.
-        if (merged !== undefined)
-            out.push(merged);
-    };
-    for (const identity of order)
-        pushMerged(groups.get(identity) ?? []);
-    for (const entry of unkeyed)
-        pushMerged([entry]);
-    // Attribute the list itself to its lowest-precedence contributor, matching
-    // `append`: a keyed list is a list every tier extends, not one a tier owns.
-    stamp(context, arrays[0], key);
-    return out;
-}
-function keyIdentity(member, field) {
-    if (!isConfigObject(member))
-        return undefined;
-    const value = member[field];
-    if (typeof value === "string")
-        return `s:${value}`;
-    if (typeof value === "number" && Number.isFinite(value)) {
-        return `n:${String(value)}`;
-    }
-    return undefined;
-}