ucn 5.5.2 → 5.6.0

package/core/deletion.js

@@ -0,0 +1,401 @@
+/**
+ * core/deletion.js - Can this declaration be deleted?
+ *
+ * One eligibility rule answers that for every surface: the dead-code
+ * pipeline (core/deadcode.js). The audit runs it over every name; an
+ * assessment runs the SAME code over one name and reads why a declaration
+ * was or was not claimed. A declaration the audit lists is exactly a
+ * declaration this module answers `no-static-dependents`.
+ *
+ * The answer is a statement of static evidence with its basis, never a
+ * promise that deletion is safe:
+ *
+ *   no-static-dependents   the name has no use in project source outside its
+ *                          own declarations, and no shield applies: nothing
+ *                          the engine can see depends on it;
+ *   has-static-dependents  a call or other use of the name exists OUTSIDE
+ *                          the declaration. The first such site is given.
+ *                          Evidence is by name, as in the audit: with two
+ *                          declarations of one name it does not say which
+ *                          of them the use reaches;
+ *   review                 no use outside the declaration was found, but a
+ *                          zero is not evidence here: a shield applies
+ *                          (consumers outside the repository, a contract, a
+ *                          registration, reflection, source that could not
+ *                          be searched) or the name is used inside the
+ *                          declaration itself. The reason is named.
+ *
+ * A use inside the declaration is deleted with it and is never a dependent:
+ * `fn len(&self) { self.inner.len() }` and an override that calls `super()`
+ * do not depend on themselves.
+ *
+ * Two things that look alike are kept apart:
+ *
+ *   downstream exposure  POLICY. An exported symbol may have consumers
+ *                        outside the repository. It blocks a claim in a
+ *                        library and not in a project its `.ucn.json`
+ *                        declares an application (or under
+ *                        --include-exported).
+ *   external contract    FACT about the declaration: it overrides or
+ *                        implements something defined outside the project,
+ *                        is registered, reflected or dispatched by name. It
+ *                        holds for unexported symbols and in applications.
+ */
+
+'use strict';
+
+const path = require('path');
+
+/** The kind a project declares for itself. Only `application` changes
+ *  anything; everything else (absent, unknown) is a library. */
+function projectKind(index) {
+    return index && index.config && index.config.kind === 'application'
+        ? 'application' : 'library';
+}
+
+/**
+ * Does the possibility of consumers outside the repository keep this
+ * exported symbol out of a deletion claim? `isExported` is the fact, read
+ * from the declaration; this is the policy over it.
+ */
+function downstreamExposureBlocks(index, isExported, options = {}) {
+    return !!isExported && !options.includeExported && projectKind(index) === 'library';
+}
+
+/** A use of the name exists: the declaration has dependents. */
+const DEPENDENT_REASONS = new Set(['has-callers', 'has-usages', 'member-in-use']);
+
+/** Why the pipeline did or did not claim a declaration, in one sentence. */
+const REASON_TEXT = Object.freeze({
+    'claimed': 'the name has no use in project source outside its own declarations, and no shield applies',
+    'has-callers': 'a call of this name exists outside the declaration',
+    'has-usages': 'the name is used outside the declaration',
+    'member-in-use': 'a member of this type is called outside the declaration',
+    'used-only-within-itself': 'every use found is inside the declaration itself, and the audit does not clear a name that is in use',
+    'downstream-exposure': 'exported from a library: consumers outside the repository may use it',
+    'external-contract': 'implements or overrides something defined outside the project',
+    'required-slot': 'required by a contract of the project: removing it breaks the type that implements it',
+    'structural-contract': 'its type has every member a project interface or protocol requires, so it can be used as one without naming it',
+    'owner-ambiguous': 'several types of its owner\'s name share the file, so its contracts cannot be checked',
+    'runtime-contract': 'invoked by the language runtime, the compiler or an assignment, not by name',
+    'registered': 'decorated or registered with a framework, which may invoke it without naming it',
+    'entry-point': 'an entry point the runtime or a test runner invokes',
+    'constructor': 'a constructor: part of its type\'s allocation protocol, assessed with the type',
+    'reflection': 'reflection in the project can reach this name',
+    'dynamic-dispatch': 'member of a registry that is called through a computed key',
+    'macro-reference': 'a macro expansion names it',
+    'unanalyzed-occurrence': 'the name occurs in source that is searched as text and not analyzed',
+    'unsearchable-source': 'part of the project could not be searched, so no name can be cleared',
+    'test-file': 'declared in a test file, which the audit does not cover by default',
+    'out-of-scope': 'outside the requested file scope',
+    'kind-not-audited': 'this kind of declaration is not audited for deletion',
+    'expression-scoped-name': 'a named function expression: its position invokes it, not its name',
+    'macro-generated': 'generated by a macro invocation, not a declaration of its own',
+    'bundled-file': 'declared in a bundled file',
+    'operator-surface': 'invoked by operator or indexer syntax that never contains its name',
+    'ambient-declaration': 'an ambient declaration that describes code outside the project',
+    'declaration-not-found': 'no declaration matches this handle',
+});
+
+function verdictOf(reason) {
+    if (reason === 'claimed') return 'no-static-dependents';
+    if (DEPENDENT_REASONS.has(reason)) return 'has-static-dependents';
+    return 'review';
+}
+
+function sameDeclaration(symbol, declaration) {
+    return symbol.name === declaration.name &&
+        symbol.file === declaration.file &&
+        symbol.startLine === declaration.startLine;
+}
+
+function relativeOf(index, file) {
+    return path.relative(index.root, file).split(path.sep).join('/');
+}
+
+/** Code-unit order, never locale order: the first site must not depend on
+ *  the machine that prints it. */
+function compareSites(a, b) {
+    if (a.file !== b.file) return a.file < b.file ? -1 : 1;
+    return (a.line || 0) - (b.line || 0);
+}
+
+/** Is this site inside the declaration, and so deleted with it? Only a
+ *  declaration the index holds has lines in the current text. */
+function withinDeclaration(declaration, indexed, file, line) {
+    if (!indexed || file !== declaration.file) return false;
+    const end = declaration.endLine != null ? declaration.endLine : declaration.startLine;
+    return line >= declaration.startLine && line <= end;
+}
+
+/**
+ * The first call of any of `names` outside the declaration, in
+ * (path, line) order, or null. A file whose call records cannot be read
+ * counts as a site with no line: an unknown is never read as "inside".
+ */
+function callSiteOutside(index, declaration, indexed, names) {
+    if (!index.calleeIndex) return null;
+    const { getCachedCalls } = require('./callers');
+    const files = new Map(); // relative path -> absolute path
+    for (const name of names) {
+        for (const file of index.calleeIndex.get(name) || []) files.set(relativeOf(index, file), file);
+    }
+    const wanted = new Set(names);
+    for (const relativePath of [...files.keys()].sort((a, b) => (a < b ? -1 : a > b ? 1 : 0))) {
+        const file = files.get(relativePath);
+        const calls = getCachedCalls(index, file);
+        if (!calls) return { file: relativePath };
+        let line = null;
+        for (const call of calls) {
+            if (!wanted.has(call.name) && !wanted.has(call.resolvedName) &&
+                !(call.resolvedNames && call.resolvedNames.some(name => wanted.has(name)))) continue;
+            if (withinDeclaration(declaration, indexed, file, call.line)) continue;
+            if (line === null || call.line < line) line = call.line;
+        }
+        if (line !== null) return { file: relativePath, line };
+    }
+    return null;
+}
+
+/**
+ * The first use of the name outside the declaration, with the pipeline's
+ * reason for it, or null.
+ *
+ * The pipeline stops at the first thing that keeps a declaration out of the
+ * audit, which is right for an audit and wrong for a verdict in two ways.
+ * A shield hides what lies behind it: an exported symbol is never scanned
+ * for uses in a library, and "consumers outside may use it" must not hide
+ * that code inside the repository does. And a call of the name anywhere
+ * ends the chain, including a call inside the declaration itself.
+ *
+ * So calls are read from the callee index directly, and other uses come
+ * from one more pass with every shield a flag can lift lifted. When no call
+ * lies outside the declaration, that pass sends the name through the text
+ * scan under the pipeline's recursion carve-out, so the pipeline's own
+ * usage rule still decides what a use is.
+ */
+function useOutsideDeclaration(index, declaration, indexed, deadcode, options) {
+    const name = declaration.name;
+    const call = callSiteOutside(index, declaration, indexed, [name]);
+    if (call) return { reason: 'has-callers', site: call };
+    let found = null;
+    deadcode(index, {
+        names: new Set([name]),
+        includeExported: true,
+        includeDecorated: true,
+        includeTests: true,
+        ownBodyOnly: new Set([name]),
+        ...(!indexed && {
+            declarations: [{ symbol: declaration, fileEntry: options.fileEntry || null }],
+        }),
+        explain: (symbol, reason, detail) => {
+            if (!sameDeclaration(symbol, declaration)) return;
+            if (reason === 'member-in-use') {
+                const site = callSiteOutside(index, declaration, indexed, (detail && detail.members) || []);
+                if (site) found = { reason, site };
+            } else if (reason === 'has-usages') {
+                const sites = ((detail && detail.usages) || [])
+                    .filter(usage => !withinDeclaration(declaration, indexed, usage.file, usage.line))
+                    .map(usage => ({ file: relativeOf(index, usage.file), line: usage.line }))
+                    .sort(compareSites);
+                if (sites.length > 0) found = { reason, site: sites[0] };
+            }
+        },
+    });
+    return found;
+}
+
+/**
+ * Assess one declaration for deletion.
+ *
+ * @param {object} index - ProjectIndex of the snapshot the declaration's
+ *   USES are looked for in
+ * @param {object} declaration - a symbol record: { name, file (absolute),
+ *   startLine, ... }. One the index holds is assessed in place. One it does
+ *   not hold (deleted since, or taken from a base revision) is assessed
+ *   from the record itself; pass `fileEntry` with the file facts of the
+ *   revision it was declared in when they differ from the index's.
+ * @param {object} [options] - { includeExported, includeDecorated,
+ *   includeTests, fileEntry }
+ */
+function assessDeletion(index, declaration, options = {}) {
+    const { deadcode } = require('./deadcode');
+    const indexed = (index.symbols.get(declaration.name) || [])
+        .some(symbol => sameDeclaration(symbol, declaration));
+    const outcomes = [];
+    const audit = deadcode(index, {
+        names: new Set([declaration.name]),
+        includeExported: !!options.includeExported,
+        includeDecorated: !!options.includeDecorated,
+        includeTests: !!options.includeTests,
+        ...(!indexed && {
+            declarations: [{ symbol: declaration, fileEntry: options.fileEntry || null }],
+        }),
+        explain: (symbol, reason, row) => outcomes.push({ symbol, reason, row }),
+    });
+    const outcome = outcomes.find(entry => sameDeclaration(entry.symbol, declaration));
+    let reason = outcome ? outcome.reason : 'declaration-not-found';
+    // --include-exported LISTS a declaration bound to an external contract,
+    // tagged, for review. Listed is not cleared: the contract is a fact
+    // about the declaration and no flag removes it.
+    if (reason === 'claimed' && outcome.row && outcome.row.externalContract) {
+        reason = 'external-contract';
+    }
+    // A use outside the declaration outranks everything but a claim. A
+    // dependent reason with no such use is no dependent: the name is used
+    // only inside what would be deleted.
+    let site = null;
+    if (outcome && reason !== 'claimed') {
+        const use = useOutsideDeclaration(index, declaration, indexed, deadcode, options);
+        if (use) {
+            reason = use.reason;
+            site = use.site;
+        } else if (DEPENDENT_REASONS.has(reason)) {
+            reason = 'used-only-within-itself';
+        }
+    }
+    const relativePath = declaration.relativePath ||
+        path.relative(index.root, declaration.file).split(path.sep).join('/');
+    return {
+        declaration: {
+            name: declaration.name,
+            file: relativePath,
+            startLine: declaration.startLine,
+            ...(declaration.endLine != null && { endLine: declaration.endLine }),
+            ...(declaration.type && { type: declaration.type }),
+            ...(declaration.className && { className: declaration.className }),
+        },
+        verdict: verdictOf(reason),
+        reason,
+        basis: REASON_TEXT[reason] || reason,
+        ...(site && { site }),
+        projectKind: projectKind(index),
+        indexed,
+        coverage: {
+            complete: audit.coverage.complete,
+            claimsWithdrawn: audit.coverage.claimsWithdrawn,
+        },
+        // Static evidence only. Aliases, reflection the engine does not
+        // model, generated code and consumers outside the repository are
+        // beyond it: the toolchain and the tests decide.
+        safeToDelete: false,
+        semanticComplete: false,
+        requiresCompilerAndTests: true,
+    };
+}
+
+/**
+ * Dependents of a declaration that CANNOT survive its deletion: sites the
+ * toolchain will reject, not sites that might be affected.
+ *
+ *   import  another file imports the name from the module that declared
+ *           it, and that module no longer declares or exports a name like
+ *           it;
+ *   call    the declaring file (or, where a directory is the package, a
+ *           file beside it) still calls the name bare, and no declaration,
+ *           import or parameter of that name remains for the call to mean.
+ *
+ * A method call on a receiver, a call through an alias, a reference by a
+ * module path: those are dependents the engine reports, never definite
+ * ones. Meant for a declaration the index no longer holds; one it holds
+ * still satisfies every site, so the answer is empty.
+ *
+ * @returns {{file, line, kind: 'import'|'call'}[]} sorted (file, line)
+ */
+/** Is `module.name` itself a module the importer resolved (a submodule)? */
+function submoduleOf(resolved, module, name) {
+    const spelled = String(module || '');
+    return !!resolved[spelled.endsWith('.') ? spelled + name : `${spelled}.${name}`];
+}
+
+function definiteDependentsOf(index, declaration) {
+    const name = declaration.name;
+    const declaredIn = declaration.relativePath || relativeOf(index, declaration.file);
+    // Declarations of the name that a module-scope reference can mean: a
+    // member, or a function local to another function, supplies nothing
+    // there.
+    const stillDeclared = (index.symbols.get(name) || []).filter(symbol =>
+        !symbol.className && symbol.lexicalScopeStartLine == null && !symbol.isNested);
+    const sites = [];
+    // Imports of the name from its former module. A module that still has
+    // a declaration of the name, still lists it among its exports, or
+    // re-exports everything from elsewhere may still provide it.
+    const formerModule = index.files.get(declaration.file);
+    const moduleStillProvides = !!formerModule && (
+        stillDeclared.some(symbol => symbol.file === declaration.file) ||
+        (formerModule.exports || []).includes(name) ||
+        (formerModule.exportDetails || []).some(detail => detail && (detail.name === '*' || detail.type === 'star')));
+    if (!moduleStillProvides) {
+        for (const entry of index.files.values()) {
+            const resolved = entry.moduleResolved || {};
+            for (const binding of entry.importBindings || []) {
+                if (binding.name !== name || binding.dynamic) continue;
+                if (resolved[binding.module] !== declaredIn) continue;
+                // `from pkg import name` where `pkg/name` is a module of
+                // the package imports that module, not a name of pkg.
+                if (submoduleOf(resolved, binding.module, name)) continue;
+                sites.push(Object.defineProperty({ file: entry.relativePath, line: binding.line || 0, kind: 'import' },
+                    'module', { value: binding.module, enumerable: false }));
+            }
+        }
+    }
+    // Bare calls with nothing left to mean. Only a free function: a member
+    // is called through a receiver. The language's own scoping decides
+    // what a bare name means at each call (languages/lexical-scope.js):
+    // only a name that resolves at module scope, in a file where no import
+    // and no other binding supplies it, is left without a meaning.
+    if (!declaration.className) {
+        const { langTraits } = require('../languages');
+        const { referenceScope } = require('../languages/lexical-scope');
+        const { parseRevision } = require('./change-kinds');
+        const traits = langTraits(formerModule?.language || index.languageOf?.(declaration.file)) || {};
+        const directoryPackage = traits.packageScope === 'directory';
+        const beside = file => file === declaration.file ||
+            (directoryPackage && path.dirname(file) === path.dirname(declaration.file));
+        // What a module-scope name can still mean in `file`: a declaration
+        // in the file, in its package directory where that is the scope,
+        // or anywhere at all where files see each other through textual
+        // inclusion.
+        // A declaration that describes code outside the project (an
+        // ambient one) supplies the name to every file.
+        const ambient = symbol => !!symbol.ambientAugmentation || /\.d\.[cm]?ts$/.test(symbol.relativePath || '');
+        const suppliedTo = file => stillDeclared.some(symbol => traits.textualIncludes || ambient(symbol) ||
+            symbol.file === file || (directoryPackage && path.dirname(symbol.file) === path.dirname(file)));
+        // A name the language itself supplies always means something.
+        const builtin = !!(traits.ambientCallables && traits.ambientCallables.has(name));
+        for (const file of builtin ? [] : [...(index.calleeIndex?.get(name) || [])].sort()) {
+            if (!beside(file) || suppliedTo(file)) continue;
+            const entry = index.files.get(file);
+            if (!entry) continue;
+            if ((entry.importBindings || []).some(binding => (binding.alias || binding.name) === name)) continue;
+            // An import of everything a module has may bring the name.
+            if ((entry.importDetails || []).some(detail => (detail.names || []).includes('*'))) continue;
+            let root;
+            try { root = parseRevision(index._readFile(file), entry.language); } catch (_) { root = null; }
+            if (!root) continue;
+            const memo = new Map();
+            const lines = new Set();
+            for (const node of root.descendantsOfType('identifier')) {
+                if (node.text !== name) continue;
+                const callee = node.parent?.childForFieldName('function');
+                if (!callee || callee.id !== node.id) continue;
+                let scope;
+                try { scope = referenceScope(node, entry.language, memo); } catch (_) { scope = null; }
+                if (scope === 'module') lines.add(node.startPosition.row + 1);
+            }
+            for (const line of lines) sites.push({ file: entry.relativePath, line, kind: 'call' });
+        }
+    }
+    return sites.sort(compareSites);
+}
+
+module.exports = {
+    definiteDependentsOf,
+    submoduleOf,
+    projectKind,
+    downstreamExposureBlocks,
+    assessDeletion,
+    verdictOf,
+    REASON_TEXT,
+    DEPENDENT_REASONS,
+};

package/core/discovery.js

@@ -8,6 +8,7 @@ const fs = require('fs');
 const path = require('path');
 const { execFileSync } = require('child_process');
 const { langTraits } = require('../languages');
+const { paramDefault } = require('./registry');
 const { IgnoreRules } = require('./ignore-rules');
 
 // Always ignore - unambiguous, never user code
@@ -61,6 +62,17 @@ const DEFAULT_IGNORES = [
 // These can contain user code: disclose their exclusion, unlike dependency
 // and VCS directories. Source maps are data and cannot be parsed as code.
 const BUNDLED_PATTERNS = ['*.min.js', '*.bundle.js', '*.map'];
+let _defaultPolicy = null;
+function defaultPolicy() {
+    if (!_defaultPolicy) _defaultPolicy = compileIgnoreList(DEFAULT_IGNORES);
+    return _defaultPolicy;
+}
+
+// Source files under paths the project's config excludes from indexing are
+// still listed, so a name can be searched in them. Past this many files the
+// excluded root is reported as one unscanned gap instead.
+const MAX_TEXT_ONLY_FILES = 20000;
+
 let _bundledList = null;
 function bundledList() {
     if (!_bundledList) _bundledList = compileIgnoreList(BUNDLED_PATTERNS);
@@ -304,14 +316,14 @@ function compareNames(a, b) {
  * @param {string} options.root - Root directory (defaults to cwd)
  * @param {string[]} options.ignores - Patterns to ignore
  * @param {number} options.maxDepth - Maximum directory depth (default: 20)
- * @param {number} options.maxFiles - Maximum files to return (default: 10000)
+ * @param {number} options.maxFiles - Maximum files to return (default: the parameter table's)
  * @returns {string[]} - Array of absolute file paths
  */
 function expandGlob(pattern, options = {}) {
     const root = path.resolve(options.root || process.cwd());
     const ignores = options.ignores || DEFAULT_IGNORES;
     const maxDepth = options.maxDepth ?? 1000;
-    const maxFiles = options.maxFiles ?? 50000;
+    const maxFiles = options.maxFiles ?? paramDefault('maxFiles');
     const maxFileSize = options.maxFileSize ?? (8 * 1024 * 1024);
     const followSymlinks = options.followSymlinks !== false; // default true
     const gitignoreRules = options.gitignoreRules || null;
@@ -345,6 +357,8 @@ function expandGlob(pattern, options = {}) {
         onDiscoveryIssue: options.onDiscoveryIssue,
         disclosureIgnores: options.disclosureIgnores,
         disclosure: options.disclosureIgnores?.length ? compileIgnoreList(options.disclosureIgnores) : null,
+        outside: options.outsideIgnores?.length ? compileIgnoreList(options.outsideIgnores) : null,
+        textOnlyState: { files: 0 },
         onFile: (filePath) => {
             if (files.length < maxFiles) {
                 files.push(filePath);
@@ -580,6 +594,7 @@ function walkDir(dir, options, depth = 0, visited = new Set(), knownRealDir = nu
         context = walkContext(options.anchorRoot || dir, dir, options.gitignoreRules || null);
     }
     const policy = options.policy || (options.policy = compileIgnoreList(options.ignores || DEFAULT_IGNORES));
+    const customIgnores = !!options.ignores && options.ignores !== DEFAULT_IGNORES;
     const rules = context.inside ? options.gitignoreRules || null : null;
     const trackedFiles = options.trackedPaths?.files;
     const trackedDirectories = options.trackedPaths?.directories;
@@ -592,14 +607,19 @@ function walkDir(dir, options, depth = 0, visited = new Set(), knownRealDir = nu
         const relative = context.rel ? `${context.rel}/${name}` : name;
         const policyIgnored = policy.matches(name, relative, entryIsDirectory) ||
             conditionallyIgnored(name, dir);
+        // Without project ignores the policy IS the built-in list: one match.
+        const defaultIgnored = customIgnores
+            ? defaultPolicy().matches(name, relative, entryIsDirectory) ||
+                conditionallyIgnored(name, dir)
+            : policyIgnored;
         // .gitignore verdict for the untracked view of this entry. Tracked
         // paths are never excluded by rules (git keeps tracking them), but a
         // tracked DIRECTORY still needs the verdict: its untracked contents
         // inherit an exclusion (git cannot re-include below an excluded
-        // directory).
+        // directory). An entry the built-in ignores already drop needs none.
         let tracked = false;
         let excluded = false;
-        if (!policyIgnored && context.inside) {
+        if (context.inside && !(policyIgnored && defaultIgnored)) {
             tracked = entryIsDirectory
                 ? !!trackedDirectories?.has(relative)
                 : !!trackedFiles?.has(relative);
@@ -608,17 +628,33 @@ function walkDir(dir, options, depth = 0, visited = new Set(), knownRealDir = nu
                 excluded = rules.decide(context.scopes, relative, context.segments, name, entryIsDirectory);
             }
         }
-        if (policyIgnored || (excluded && !tracked)) {
-            if (options.disclosure && options.disclosure.matches(name, relative, entryIsDirectory)) {
+        const gitExcluded = excluded && !tracked;
+        // Why an entry is not indexed decides what it means for a claim:
+        //   declared outside the project   no part of any claim, stated once;
+        //   built-in ignores, .gitignore   no part of any claim (as ever);
+        //   excluded by config ALONE       "do not index": its source is
+        //                                  still listed so a name can be
+        //                                  searched in it (text-only).
+        let textOnly = context.textOnly || null;
+        if (policyIgnored || gitExcluded || textOnly) {
+            if (options.outside && options.outside.matches(name, relative, entryIsDirectory)) {
                 options.onDiscoveryIssue?.({
                     path: fullPath,
                     kind: entryIsDirectory ? 'directory' : 'file',
-                    reason: 'config-exclude',
-                    detail: 'excluded by .ucn.json',
+                    reason: 'outside-project',
+                    detail: 'declared outside the project by .ucn.json',
                 });
+                continue;
+            }
+            if (defaultIgnored || gitExcluded) continue;
+            if (!textOnly) {
+                const configOnly = options.disclosure &&
+                    options.disclosure.matches(name, relative, entryIsDirectory);
+                if (!configOnly) continue;
+                textOnly = 'config-exclude';
             }
-            continue;
         }
+        const textOnlyRoot = textOnly ? (context.textOnlyRoot || fullPath) : null;
 
         let isDir = entry.isDirectory();
         let isFile = entry.isFile();
@@ -643,8 +679,32 @@ function walkDir(dir, options, depth = 0, visited = new Set(), knownRealDir = nu
                         segments: context.segments + 1,
                         scopes: rules && !excluded ? rules.scopesFor(context.scopes, relative) : context.scopes,
                         excluded,
+                        textOnly,
+                        textOnlyRoot,
                     });
             }
+        } else if (isFile && textOnly) {
+            // Listed, never indexed: only what discovery would have treated
+            // as source (a supported or known unsupported language, a
+            // bundle). Source maps and other data stay out, as everywhere.
+            const source = options.filePattern.test(entry.name) ||
+                !!classifyUnsupportedSourceFile(fullPath) ||
+                (bundledList().matches(entry.name, null, false) && !entry.name.endsWith('.map'));
+            if (!source) continue;
+            const state = options.textOnlyState || (options.textOnlyState = { files: 0 });
+            if (state.files >= MAX_TEXT_ONLY_FILES) {
+                options.onDiscoveryIssue?.({
+                    path: textOnlyRoot, kind: 'directory', reason: 'text-scan-budget',
+                    detail: `more than ${MAX_TEXT_ONLY_FILES} source files under paths ` +
+                        'excluded by .ucn.json; declare vendored trees in outsideProject',
+                });
+                continue;
+            }
+            state.files++;
+            options.onDiscoveryIssue?.({
+                path: fullPath, kind: 'file', reason: textOnly,
+                detail: 'excluded by .ucn.json',
+            });
         } else if (isFile) {
             if (bundledList().matches(entry.name, null, false) &&
                 (!options.includeBundled || entry.name.endsWith('.map'))) {
@@ -1008,8 +1068,43 @@ function findTestFileFor(sourceFile, language) {
     return null;
 }
 
+function patternList(value) {
+    if (Array.isArray(value)) return value.map(String);
+    return value ? [String(value)] : [];
+}
+
+/**
+ * The discovery options a project's configuration implies. The build and
+ * the staleness walk both start here: a walk that reads the config
+ * differently sees a different file set and calls a fresh index stale (or a
+ * stale one fresh).
+ *
+ * @param {object} config - parsed .ucn.json (may be empty)
+ * @param {object} [overrides] - explicit maxFiles / maxDepth / maxFileSize /
+ *   followSymlinks / includeBundled; undefined keeps the config or default
+ */
+function projectDiscoveryOptions(root, config = {}, overrides = {}) {
+    const exclude = patternList(config.exclude);
+    const outsideProject = patternList(config.outsideProject);
+    const options = {
+        root,
+        maxFiles: overrides.maxFiles || config.maxFiles || paramDefault('maxFiles'),
+        maxDepth: overrides.maxDepth ?? config.maxDepth,
+        maxFileSize: overrides.maxFileSize ?? config.maxFileSize,
+        followSymlinks: overrides.followSymlinks,
+        includeBundled: overrides.includeBundled === true,
+        disclosureIgnores: exclude,
+        outsideIgnores: outsideProject,
+    };
+    if (exclude.length > 0 || outsideProject.length > 0) {
+        options.ignores = [...DEFAULT_IGNORES, ...exclude, ...outsideProject];
+    }
+    return options;
+}
+
 module.exports = {
     expandGlob,
+    projectDiscoveryOptions,
     parseGlobPattern,
     globToRegex,
     walkDir,

package/core/entrypoints.js

@@ -595,6 +595,12 @@ const FRAMEWORK_PATTERNS = [
 
 ];
 
+// What the `type` and `framework` filters accept: the kinds and the labels a
+// pattern above carries, and nothing else.
+const ENTRYPOINT_TYPES = Object.freeze([...new Set(FRAMEWORK_PATTERNS.map(p => p.type))].sort());
+const ENTRYPOINT_FRAMEWORKS = Object.freeze(
+    [...new Set(FRAMEWORK_PATTERNS.map(p => p.framework.toLowerCase()))].sort());
+
 // ============================================================================
 // DETECTION
 // ============================================================================
@@ -864,11 +870,10 @@ function detectEntrypoints(index, options = {}) {
     // Validate --type against the pattern registry up front — an unknown
     // value used to fall through to the filter and silently return nothing.
     if (options.type) {
-        const validTypes = new Set(FRAMEWORK_PATTERNS.map(p => p.type));
-        if (!validTypes.has(options.type)) {
+        if (!ENTRYPOINT_TYPES.includes(options.type)) {
             return {
                 error: 'invalid-type',
-                message: `Unknown type "${options.type}". Valid: ${[...validTypes].sort().join(', ')}.`,
+                message: `Unknown type "${options.type}". Valid: ${ENTRYPOINT_TYPES.join(', ')}.`,
             };
         }
     }
@@ -876,13 +881,12 @@ function detectEntrypoints(index, options = {}) {
     // Same discipline for --framework (fix #243) — a typo like 'flsk'
     // silently filtered everything to an empty result.
     if (options.framework) {
-        const validFrameworks = new Set(FRAMEWORK_PATTERNS.map(p => p.framework.toLowerCase()));
         const wanted = String(options.framework).split(',').map(s => s.trim().toLowerCase()).filter(Boolean);
-        const unknown = wanted.filter(f => !validFrameworks.has(f));
+        const unknown = wanted.filter(f => !ENTRYPOINT_FRAMEWORKS.includes(f));
         if (unknown.length > 0) {
             return {
                 error: 'invalid-framework',
-                message: `Unknown framework "${unknown.join('", "')}". Valid: ${[...validFrameworks].sort().join(', ')}.`,
+                message: `Unknown framework "${unknown.join('", "')}". Valid: ${ENTRYPOINT_FRAMEWORKS.join(', ')}.`,
             };
         }
     }
@@ -1462,6 +1466,8 @@ function isFrameworkEntrypoint(symbol, index) {
 }
 
 module.exports = {
+    ENTRYPOINT_TYPES,
+    ENTRYPOINT_FRAMEWORKS,
     FRAMEWORK_PATTERNS,
     detectEntrypoints,
     isFrameworkEntrypoint,