gitnexus 1.6.8-rc.12 → 1.6.8-rc.13

package/dist/core/ingestion/cfg/visitors/typescript.d.ts

@@ -0,0 +1,49 @@
+/**
+ * TS/JS CfgVisitor (issue #2081, M1).
+ *
+ * Walks a TypeScript/JavaScript function's tree-sitter AST and drives the
+ * language-agnostic {@link CfgBuilder} to produce a serializable
+ * {@link FunctionCfg}. TS and JS share a grammar family (tree-sitter-typescript
+ * reuses tree-sitter-javascript's statement nodes), so one visitor covers both.
+ *
+ * Design — a `visit_<node_type>` dispatch over the statement taxonomy. The
+ * classic CFG hazards (R10) are handled explicitly:
+ *  - loops allocate a dedicated **loop-exit** block so `break` has a concrete
+ *    target before the loop's successor is known; `continue` targets the
+ *    header/increment; the back-edge closes the loop.
+ *  - `switch` cases fall through naturally: a case body that does not `break`
+ *    yields non-empty `exits`, which we wire to the next case as `fallthrough`;
+ *    a case that `break`s wires to the switch exit (via {@link ControlFlowContext})
+ *    and yields no fall-out.
+ *  - `try/catch/finally` routes both normal completion AND a `throw` in the try
+ *    through `finally` (the finally block post-dominates the try/catch); a
+ *    `throw` with no catch propagates through finally to the enclosing handler.
+ *  - labeled `break`/`continue` resolve against the labeled loop's frame.
+ *
+ * Known M1 limitations:
+ *  - SOUNDNESS GAP (M2 blocker, not mere precision): a non-local jump
+ *    (`break`/`continue`/`return`) out of a `try` that has a `finally` edges
+ *    directly to its target rather than routing THROUGH the `finally` block
+ *    first. A future taint/PDG pass will therefore MISS flow mediated by a
+ *    `finally` on the early-exit path (e.g. a value the `finally` taints or
+ *    sanitizes before the `return` reaches its target) — a false negative. The
+ *    general fix duplicates `finally` per exit path; deferred past M1 and
+ *    tracked for M2. Normal completion and `throw` DO route through `finally`.
+ *  - A `break`/`continue` to a label on a non-loop/non-switch block, and the
+ *    OUTER label of a doubly-labeled construct (`outer: inner: for (...)`), are
+ *    not modeled. The jump is conservatively routed to the function EXIT (a
+ *    sound over-approximation that keeps the graph single-exit — see visitBreak)
+ *    rather than left as a dangling sink; only the precise labeled target is
+ *    unmodeled. Single-labeled loops/switches resolve correctly.
+ *
+ * Block/edge accounting and reachability are pinned in
+ * `test/unit/cfg/cfg-builder.test.ts` (core) and
+ * `test/unit/cfg/typescript-visitor.test.ts` (this visitor, per hazard).
+ */
+import type { SyntaxNode } from '../../utils/ast-helpers.js';
+import type { CfgVisitor } from '../types.js';
+/** TS/JS node types that own a CFG-bearing function body. */
+declare const TS_FUNCTION_TYPES: Set<string>;
+/** The TS/JS CFG visitor (shared by TypeScript and JavaScript). */
+export declare function createTypeScriptCfgVisitor(): CfgVisitor<SyntaxNode>;
+export { TS_FUNCTION_TYPES };

package/dist/core/ingestion/cfg/visitors/typescript.js

@@ -0,0 +1,492 @@
+import { CfgBuilder } from '../cfg-builder.js';
+import { ControlFlowContext } from '../control-flow-context.js';
+/** TS/JS node types that own a CFG-bearing function body. */
+const TS_FUNCTION_TYPES = new Set([
+    'function_declaration',
+    'function_expression',
+    'arrow_function',
+    'method_definition',
+    'generator_function_declaration',
+    'generator_function',
+    'async_function_declaration',
+    'async_arrow_function',
+]);
+/** Statement node types that break a basic block (everything else coalesces). */
+const CONTROL_FLOW_TYPES = new Set([
+    'if_statement',
+    'while_statement',
+    'do_statement',
+    'for_statement',
+    'for_in_statement',
+    'for_of_statement',
+    'switch_statement',
+    'try_statement',
+    'return_statement',
+    'break_statement',
+    'continue_statement',
+    'throw_statement',
+    'labeled_statement',
+    'statement_block',
+]);
+const LOOP_OR_SWITCH_TYPES = new Set([
+    'while_statement',
+    'do_statement',
+    'for_statement',
+    'for_in_statement',
+    'for_of_statement',
+    'switch_statement',
+]);
+const startLineOf = (n) => n.startPosition.row + 1;
+const endLineOf = (n) => n.endPosition.row + 1;
+/**
+ * Per-function walk state. One instance is created per function so the
+ * {@link ControlFlowContext}, exception-handler stack, and pending label are
+ * scoped to that function and never leak across functions.
+ */
+class TsCfgWalk {
+    builder;
+    cfc = new ControlFlowContext();
+    /** Stack of exception-handler entry blocks (catch/finally) a `throw` jumps to. */
+    handlers = [];
+    /** Label awaiting the loop/switch it immediately precedes (labeled_statement). */
+    pendingLabel;
+    constructor(builder) {
+        this.builder = builder;
+    }
+    /** Statements of a block node, ignoring comments. */
+    statementsOf(block) {
+        return block.namedChildren.filter((c) => c.type !== 'comment');
+    }
+    /** The `body` block of a node (field, or the first statement_block child). */
+    bodyBlockOf(node) {
+        return (node.childForFieldName('body') ?? node.namedChildren.find((c) => c.type === 'statement_block'));
+    }
+    /** Visit a body that may be a `statement_block` or a single statement. */
+    visitBody(node) {
+        if (!node)
+            return null;
+        if (node.type === 'statement_block')
+            return this.visitSeq(this.statementsOf(node));
+        return this.visitStmt(node);
+    }
+    /** Wire a sequence of statements, coalescing straight-line runs into blocks. */
+    visitSeq(stmts) {
+        let entry;
+        let dangling = [];
+        let openSimple;
+        for (const stmt of stmts) {
+            if (CONTROL_FLOW_TYPES.has(stmt.type)) {
+                openSimple = undefined; // close any open straight-line block
+                const res = this.visitStmt(stmt);
+                if (res === null)
+                    continue; // transparent (empty nested block)
+                if (entry === undefined)
+                    entry = res.entry;
+                else
+                    this.builder.connect(dangling, res.entry, 'seq');
+                dangling = [...res.exits];
+            }
+            else {
+                // Simple statement — coalesce into the current straight-line block.
+                if (openSimple === undefined) {
+                    const idx = this.builder.newBlock(startLineOf(stmt), endLineOf(stmt), stmt.text);
+                    if (entry === undefined)
+                        entry = idx;
+                    else
+                        this.builder.connect(dangling, idx, 'seq');
+                    openSimple = idx;
+                    dangling = [idx];
+                }
+                else {
+                    this.builder.extendBlock(openSimple, endLineOf(stmt), stmt.text);
+                }
+            }
+        }
+        if (entry === undefined)
+            return null;
+        return { entry, exits: dangling };
+    }
+    /** Dispatch one statement to its handler. Non-null except for empty blocks. */
+    visitStmt(stmt) {
+        switch (stmt.type) {
+            case 'if_statement':
+                return this.visitIf(stmt);
+            case 'while_statement':
+                return this.visitWhile(stmt);
+            case 'do_statement':
+                return this.visitDoWhile(stmt);
+            case 'for_statement':
+                return this.visitFor(stmt);
+            case 'for_in_statement':
+            case 'for_of_statement':
+                return this.visitForIn(stmt);
+            case 'switch_statement':
+                return this.visitSwitch(stmt);
+            case 'try_statement':
+                return this.visitTry(stmt);
+            case 'return_statement':
+                return this.visitReturn(stmt);
+            case 'throw_statement':
+                return this.visitThrow(stmt);
+            case 'break_statement':
+                return this.visitBreak(stmt);
+            case 'continue_statement':
+                return this.visitContinue(stmt);
+            case 'labeled_statement':
+                return this.visitLabeled(stmt);
+            case 'statement_block':
+                return this.visitSeq(this.statementsOf(stmt));
+            default:
+                return this.visitSimple(stmt);
+        }
+    }
+    visitSimple(stmt) {
+        const idx = this.builder.newBlock(startLineOf(stmt), endLineOf(stmt), stmt.text);
+        return { entry: idx, exits: [idx] };
+    }
+    visitReturn(stmt) {
+        const idx = this.builder.newBlock(startLineOf(stmt), endLineOf(stmt), stmt.text);
+        this.builder.edge(idx, this.builder.exitIndex, 'return');
+        return { entry: idx, exits: [] };
+    }
+    visitThrow(stmt) {
+        const idx = this.builder.newBlock(startLineOf(stmt), endLineOf(stmt), stmt.text);
+        this.builder.edge(idx, this.currentHandler(), 'throw');
+        return { entry: idx, exits: [] };
+    }
+    visitBreak(stmt) {
+        const idx = this.builder.newBlock(startLineOf(stmt), endLineOf(stmt), stmt.text);
+        const target = this.cfc.breakTarget(this.labelOf(stmt));
+        // An unresolved target — a label this M1 visitor doesn't model (a stacked
+        // outer label like `outer: inner: for`, or a labeled non-loop block) —
+        // would otherwise leave this block with NO out-edge, stranding it and
+        // breaking the single-exit invariant a downstream post-dominator / PDG pass
+        // relies on. Conservatively route an unresolved jump to the function EXIT
+        // ("escapes the function"): sound over-approximation, keeps single-exit.
+        this.builder.edge(idx, target ?? this.builder.exitIndex, 'break');
+        return { entry: idx, exits: [] };
+    }
+    visitContinue(stmt) {
+        const idx = this.builder.newBlock(startLineOf(stmt), endLineOf(stmt), stmt.text);
+        const target = this.cfc.continueTarget(this.labelOf(stmt));
+        // See visitBreak: an unresolved label routes to EXIT to preserve single-exit.
+        this.builder.edge(idx, target ?? this.builder.exitIndex, 'continue');
+        return { entry: idx, exits: [] };
+    }
+    visitLabeled(stmt) {
+        const body = stmt.childForFieldName('body') ?? stmt.namedChildren[stmt.namedChildren.length - 1];
+        if (body && LOOP_OR_SWITCH_TYPES.has(body.type)) {
+            this.pendingLabel = this.labelOf(stmt);
+            const res = this.visitStmt(body);
+            this.pendingLabel = undefined; // clear even if the construct didn't consume it
+            return res;
+        }
+        // Labeled non-loop blocks (break-to-block-label) are not modeled in M1.
+        return this.visitBody(body);
+    }
+    visitIf(stmt) {
+        const cond = stmt.childForFieldName('condition') ?? stmt;
+        const condBlock = this.builder.newBlock(startLineOf(stmt), endLineOf(cond), cond.text);
+        const exits = [];
+        const thenRes = this.visitBody(stmt.childForFieldName('consequence'));
+        if (thenRes) {
+            this.builder.edge(condBlock, thenRes.entry, 'cond-true');
+            exits.push(...thenRes.exits);
+        }
+        else {
+            exits.push(condBlock); // empty then — true path falls through
+        }
+        const elseNode = this.elseBodyOf(stmt);
+        if (elseNode) {
+            const elseRes = this.visitBody(elseNode);
+            if (elseRes) {
+                this.builder.edge(condBlock, elseRes.entry, 'cond-false');
+                exits.push(...elseRes.exits);
+            }
+            else {
+                exits.push(condBlock); // empty else block
+            }
+        }
+        else {
+            exits.push(condBlock); // no else — false path falls through to the join
+        }
+        return { entry: condBlock, exits: [...new Set(exits)] };
+    }
+    /** The else body node (unwraps an `else_clause` wrapper if present). */
+    elseBodyOf(ifStmt) {
+        const alt = ifStmt.childForFieldName('alternative');
+        if (!alt)
+            return undefined;
+        if (alt.type === 'else_clause') {
+            return alt.childForFieldName('body') ?? alt.namedChildren[0];
+        }
+        return alt;
+    }
+    visitWhile(stmt) {
+        const label = this.takeLabel();
+        const cond = stmt.childForFieldName('condition') ?? stmt;
+        const header = this.builder.newBlock(startLineOf(stmt), endLineOf(cond), cond.text);
+        const loopExit = this.builder.newBlock(endLineOf(stmt), endLineOf(stmt), '');
+        this.cfc.pushLoop(header, loopExit, label);
+        const body = this.visitBody(this.bodyBlockOf(stmt));
+        this.cfc.pop();
+        if (body) {
+            this.builder.edge(header, body.entry, 'cond-true');
+            this.builder.connect(body.exits, header, 'loop-back');
+        }
+        else {
+            this.builder.edge(header, header, 'loop-back'); // empty body re-tests
+        }
+        this.builder.edge(header, loopExit, 'cond-false');
+        return { entry: header, exits: [loopExit] };
+    }
+    visitDoWhile(stmt) {
+        const label = this.takeLabel();
+        const cond = stmt.childForFieldName('condition') ?? stmt;
+        const condBlock = this.builder.newBlock(startLineOf(cond), endLineOf(cond), cond.text);
+        const loopExit = this.builder.newBlock(endLineOf(stmt), endLineOf(stmt), '');
+        this.cfc.pushLoop(condBlock, loopExit, label);
+        const body = this.visitBody(this.bodyBlockOf(stmt));
+        this.cfc.pop();
+        const backTarget = body ? body.entry : condBlock;
+        if (body)
+            this.builder.connect(body.exits, condBlock, 'seq');
+        this.builder.edge(condBlock, backTarget, 'loop-back'); // cond true → run body again
+        this.builder.edge(condBlock, loopExit, 'cond-false');
+        return { entry: backTarget, exits: [loopExit] };
+    }
+    visitFor(stmt) {
+        const label = this.takeLabel();
+        const init = stmt.childForFieldName('initializer');
+        const cond = stmt.childForFieldName('condition');
+        const incr = stmt.childForFieldName('increment');
+        const header = this.builder.newBlock(startLineOf(stmt), cond ? endLineOf(cond) : startLineOf(stmt), cond ? cond.text : 'for(;;)');
+        const loopExit = this.builder.newBlock(endLineOf(stmt), endLineOf(stmt), '');
+        let incrBlock = header;
+        if (incr) {
+            incrBlock = this.builder.newBlock(startLineOf(incr), endLineOf(incr), incr.text);
+            this.builder.edge(incrBlock, header, 'loop-back');
+        }
+        this.cfc.pushLoop(incrBlock, loopExit, label);
+        const body = this.visitBody(this.bodyBlockOf(stmt));
+        this.cfc.pop();
+        if (body) {
+            this.builder.edge(header, body.entry, 'cond-true');
+            // With no increment clause the body's exits ARE the back-edge — carry
+            // the loop-back kind on them (mirroring visitWhile/visitForIn) instead
+            // of a phantom header→header self-loop that models a path which never
+            // executes the body. With an increment, the body falls through to the
+            // increment (`seq`) and the increment carries the loop-back (:338).
+            this.builder.connect(body.exits, incrBlock, incr ? 'seq' : 'loop-back');
+        }
+        else {
+            this.builder.edge(header, incrBlock, 'cond-true');
+            // Empty body with no increment: the header genuinely re-tests itself.
+            if (!incr)
+                this.builder.edge(header, header, 'loop-back');
+        }
+        this.builder.edge(header, loopExit, 'cond-false');
+        let entry = header;
+        if (init) {
+            const initBlock = this.builder.newBlock(startLineOf(init), endLineOf(init), init.text);
+            this.builder.edge(initBlock, header, 'seq');
+            entry = initBlock;
+        }
+        return { entry, exits: [loopExit] };
+    }
+    visitForIn(stmt) {
+        const label = this.takeLabel();
+        const header = this.builder.newBlock(startLineOf(stmt), startLineOf(stmt), this.forInHeaderText(stmt));
+        const loopExit = this.builder.newBlock(endLineOf(stmt), endLineOf(stmt), '');
+        this.cfc.pushLoop(header, loopExit, label);
+        const body = this.visitBody(this.bodyBlockOf(stmt));
+        this.cfc.pop();
+        if (body) {
+            this.builder.edge(header, body.entry, 'cond-true');
+            this.builder.connect(body.exits, header, 'loop-back');
+        }
+        else {
+            this.builder.edge(header, header, 'loop-back');
+        }
+        this.builder.edge(header, loopExit, 'cond-false');
+        return { entry: header, exits: [loopExit] };
+    }
+    forInHeaderText(stmt) {
+        const left = stmt.childForFieldName('left')?.text ?? '';
+        const right = stmt.childForFieldName('right')?.text ?? '';
+        return left || right ? `for(${left} … ${right})` : 'for(… in/of …)';
+    }
+    visitSwitch(stmt) {
+        const label = this.takeLabel();
+        const value = stmt.childForFieldName('value') ?? stmt;
+        const dispatch = this.builder.newBlock(startLineOf(stmt), endLineOf(value), value.text);
+        const switchExit = this.builder.newBlock(endLineOf(stmt), endLineOf(stmt), '');
+        this.cfc.pushSwitch(switchExit, label);
+        const body = stmt.childForFieldName('body');
+        const cases = body
+            ? body.namedChildren.filter((c) => c.type === 'switch_case' || c.type === 'switch_default')
+            : [];
+        const caseResults = cases.map((c) => this.visitSeq(this.caseStatements(c)));
+        const hasDefault = cases.some((c) => c.type === 'switch_default');
+        // entryOf[i] = block a dispatch/fallthrough INTO case i lands on (empty
+        // cases are transparent — they resolve to the next case, or the exit).
+        const entryOf = new Array(cases.length);
+        let after = switchExit;
+        for (let i = cases.length - 1; i >= 0; i--) {
+            entryOf[i] = caseResults[i]?.entry ?? after;
+            after = entryOf[i];
+        }
+        for (let i = 0; i < cases.length; i++) {
+            this.builder.edge(dispatch, entryOf[i], 'switch-case');
+        }
+        if (!hasDefault)
+            this.builder.edge(dispatch, switchExit, 'switch-case'); // no-match path
+        for (let i = 0; i < cases.length; i++) {
+            const res = caseResults[i];
+            if (!res)
+                continue;
+            const fallTarget = i + 1 < cases.length ? entryOf[i + 1] : switchExit;
+            this.builder.connect(res.exits, fallTarget, 'fallthrough');
+        }
+        this.cfc.pop();
+        return { entry: dispatch, exits: [switchExit] };
+    }
+    caseStatements(caseNode) {
+        const value = caseNode.childForFieldName('value');
+        return caseNode.namedChildren.filter((c) => c.id !== value?.id && c.type !== 'comment');
+    }
+    visitTry(stmt) {
+        const bodyNode = stmt.childForFieldName('body');
+        // Single pass over named children — tree-sitter's `namedChildren` getter
+        // allocates a fresh array on every access, so avoid the double `.find`.
+        let catchClause;
+        let finallyClause;
+        for (let i = 0; i < stmt.namedChildCount; i++) {
+            const c = stmt.namedChild(i);
+            if (c?.type === 'catch_clause')
+                catchClause = c;
+            else if (c?.type === 'finally_clause')
+                finallyClause = c;
+        }
+        // Build finally first so its entry is known as both a normal join and a
+        // handler target. The finally body runs in the OUTER handler context.
+        const finallyRes = finallyClause
+            ? this.visitSeq(this.statementsOf(this.bodyBlockOf(finallyClause)))
+            : null;
+        // A throw inside catch propagates to finally (if any), else the outer handler.
+        let catchRes = null;
+        if (catchClause) {
+            if (finallyRes)
+                this.handlers.push(finallyRes.entry);
+            catchRes = this.visitSeq(this.statementsOf(this.bodyBlockOf(catchClause)));
+            if (finallyRes)
+                this.handlers.pop();
+            if (catchRes === null) {
+                // Empty (or comment-only) catch body — `catch {}`. The clause still
+                // CATCHES: handler semantics key off the syntactic clause, not the
+                // traversal result. Treating it as "no catch" sent the swallowed
+                // exception to the outer handler/EXIT and left post-try code
+                // unreachable when the body always throws — a hard false-negative
+                // for downstream taint. Synthesize one empty block spanning the
+                // clause (entry == sole exit) so exception flow lands in it and
+                // rejoins the normal continuation. Created BEFORE the protected
+                // region is walked, so it never receives a spurious throw edge.
+                const idx = this.builder.newBlock(startLineOf(catchClause), endLineOf(catchClause), '');
+                catchRes = { entry: idx, exits: [idx] };
+            }
+        }
+        // Handler for the try body: catch if present, else finally, else outer.
+        const tryHandler = catchRes?.entry ?? finallyRes?.entry ?? this.currentHandler();
+        const protectedStart = this.builder.blockCount;
+        this.handlers.push(tryHandler);
+        const bodyRes = bodyNode ? this.visitSeq(this.statementsOf(bodyNode)) : null;
+        this.handlers.pop();
+        // Conservative exceptional edges: ANY block in the protected region may raise
+        // to the handler — not just an explicit `throw`, and not just the body ENTRY.
+        // Edging every block created during the try-body walk keeps exception flow
+        // sound when the body BRANCHES: an `if` / nested-try / post-branch block whose
+        // interior blocks would otherwise have no path to the handler — i.e. a taint
+        // false-negative into `catch` for the downstream PDG analysis. The
+        // per-function edge cap bounds the count; explicit `throw`s add their own
+        // (idempotent) edge to the same handler.
+        if (catchClause || finallyClause) {
+            for (let b = protectedStart; b < this.builder.blockCount; b++) {
+                this.builder.edge(b, tryHandler, 'throw');
+            }
+        }
+        const exits = [];
+        if (finallyRes) {
+            // Normal completion of try AND catch both flow through finally.
+            if (bodyRes)
+                this.builder.connect(bodyRes.exits, finallyRes.entry, 'seq');
+            if (catchRes)
+                this.builder.connect(catchRes.exits, finallyRes.entry, 'seq');
+            exits.push(...finallyRes.exits);
+            // No catch → an exception re-propagates out after finally runs.
+            if (!catchRes)
+                this.builder.connect(finallyRes.exits, this.currentHandler(), 'throw');
+        }
+        else {
+            if (bodyRes)
+                exits.push(...bodyRes.exits);
+            if (catchRes)
+                exits.push(...catchRes.exits);
+        }
+        const entry = bodyRes?.entry ?? finallyRes?.entry ?? catchRes?.entry;
+        if (entry === undefined)
+            return null;
+        return { entry, exits: [...new Set(exits)] };
+    }
+    /** Nearest enclosing exception handler, or the function EXIT. */
+    currentHandler() {
+        return this.handlers.length ? this.handlers[this.handlers.length - 1] : this.builder.exitIndex;
+    }
+    /** Consume the label awaiting the loop/switch this call is building. */
+    takeLabel() {
+        const label = this.pendingLabel;
+        this.pendingLabel = undefined;
+        return label;
+    }
+    labelOf(stmt) {
+        const id = stmt.childForFieldName('label') ??
+            stmt.namedChildren.find((c) => c.type === 'statement_identifier');
+        return id?.text;
+    }
+}
+/** Build the CFG for one TS/JS function node (or `undefined` if not a function). */
+function buildFunctionCfg(fnNode, filePath) {
+    if (!TS_FUNCTION_TYPES.has(fnNode.type))
+        return undefined;
+    const startLine = startLineOf(fnNode);
+    const endLine = endLineOf(fnNode);
+    const startColumn = fnNode.startPosition.column;
+    const builder = new CfgBuilder(filePath, startLine, endLine, startColumn);
+    const body = fnNode.childForFieldName('body');
+    if (!body)
+        return undefined; // overload signature / abstract method — no body
+    if (body.type !== 'statement_block') {
+        // Expression-bodied arrow: `() => expr` — one block whose value is returned.
+        const blk = builder.newBlock(startLineOf(body), endLineOf(body), body.text);
+        builder.edge(builder.entryIndex, blk, 'seq');
+        builder.edge(blk, builder.exitIndex, 'return');
+        return builder.finish();
+    }
+    const walk = new TsCfgWalk(builder);
+    const res = walk.visitSeq(body.namedChildren.filter((c) => c.type !== 'comment'));
+    if (!res) {
+        builder.edge(builder.entryIndex, builder.exitIndex, 'seq'); // empty body
+        return builder.finish();
+    }
+    builder.edge(builder.entryIndex, res.entry, 'seq');
+    builder.connect(res.exits, builder.exitIndex, 'seq'); // normal fall-off → EXIT
+    return builder.finish();
+}
+/** Whether a node is a TS/JS function this visitor builds a CFG for. */
+function isFunction(node) {
+    return TS_FUNCTION_TYPES.has(node.type);
+}
+/** The TS/JS CFG visitor (shared by TypeScript and JavaScript). */
+export function createTypeScriptCfgVisitor() {
+    return { buildFunctionCfg, isFunction };
+}
+export { TS_FUNCTION_TYPES };

package/dist/core/ingestion/language-provider.d.ts

@@ -19,6 +19,7 @@ import type { MethodExtractor } from './method-types.js';
 import type { VariableExtractor } from './variable-types.js';
 import type { ImportResolverFn } from './import-resolvers/types.js';
 import type { SyntaxNode } from './utils/ast-helpers.js';
+import type { CfgVisitor } from './cfg/types.js';
 import type { NodeLabel } from '../../_shared/index.js';
 import type Parser from 'tree-sitter';
 import type { ExtractedDecoratorRoute } from './workers/parse-worker.js';
@@ -284,6 +285,16 @@ interface LanguageProviderConfig {
      * Default: undefined (provider has no capture-time module-level side effects).
      */
     readonly collectCaptureSideChannel?: (filePath: string) => unknown;
+    /**
+     * Per-language control-flow-graph builder (#2081 M1, PDG/taint substrate).
+     * Invoked IN THE PARSE WORKER (where the AST lives) for each function node,
+     * gated on the `--pdg` opt-in; the resulting per-function CFGs are serialized
+     * onto `ParsedFile.cfgSideChannel` and emitted as BasicBlock nodes + CFG
+     * edges during scope-resolution. `TNode` is `SyntaxNode` for the tree-sitter
+     * languages. Default: undefined (language has no CFG support yet — TS/JS are
+     * the M1 set).
+     */
+    readonly cfgVisitor?: CfgVisitor<SyntaxNode>;
     /**
      * Interpret a raw `@import.statement` capture group into a `ParsedImport`.
      * The central finalize algorithm resolves `ParsedImport.targetRaw` to a

package/dist/core/ingestion/languages/typescript.js

@@ -9,6 +9,7 @@ import { SupportedLanguages } from '../../../_shared/index.js';
 import { defineLanguage } from '../language-provider.js';
 import { createClassExtractor } from '../class-extractors/generic.js';
 import { typescriptClassConfig, javascriptClassConfig, } from '../class-extractors/configs/typescript-javascript.js';
+import { createTypeScriptCfgVisitor } from '../cfg/visitors/typescript.js';
 import { typeConfig as typescriptConfig } from '../type-extractors/typescript.js';
 import { tsExportChecker } from '../export-detection.js';
 import { createImportResolver } from '../import-resolvers/resolver-factory.js';
@@ -293,6 +294,8 @@ export const typescriptProvider = defineLanguage({
     // canonical capture vocabulary in ./typescript/query.ts
     // (TYPESCRIPT_SCOPE_QUERY constant).
     emitScopeCaptures: emitTsScopeCaptures,
+    // CFG/PDG substrate (#2081 M1) — runs in the worker on a --pdg run.
+    cfgVisitor: createTypeScriptCfgVisitor(),
     interpretImport: interpretTsImport,
     interpretTypeBinding: interpretTsTypeBinding,
     bindingScopeFor: tsBindingScopeFor,
@@ -352,6 +355,8 @@ export const javascriptProvider = defineLanguage({
     // JSDoc type bindings) live in ./javascript/captures.ts.
     // See ./javascript/index.ts for the full per-module rationale.
     emitScopeCaptures: emitJsScopeCaptures,
+    // CFG/PDG substrate (#2081 M1) — TS and JS share the same grammar family.
+    cfgVisitor: createTypeScriptCfgVisitor(),
     interpretImport: interpretJsImport,
     interpretTypeBinding: interpretJsTypeBinding,
     bindingScopeFor: jsBindingScopeFor,

package/dist/core/ingestion/pipeline-phases/parse-impl.js

@@ -18,6 +18,7 @@ import { BindingAccumulator, enrichExportedTypeMap, } from '../binding-accumulat
 import { mergeChunkResults, dispatchChunkParse } from '../parsing-processor.js';
 import { fileContentHash, computeChunkHash, loadParseCacheChunk, persistParseCacheChunk, PARSE_CACHE_VERSION, } from '../../../storage/parse-cache.js';
 import { clearParsedFileStore, persistParsedFileChunk, getDurableParsedFileDir, loadDurableParsedFileIndex, restoreDurableParsedFileShard, } from '../../../storage/parsedfile-store.js';
+import { DEFAULT_PDG_MAX_FUNCTION_LINES } from '../cfg/collect.js';
 import { processRoutesFromExtracted, buildExportedTypeMapFromGraph, } from '../call-processor.js';
 import { createSemanticModel } from '../model/index.js';
 import { getLanguageFromFilename, } from '../../../_shared/index.js';
@@ -319,6 +320,10 @@ export async function runChunkedParseAndResolve(graph, scannedFiles, allPaths, t
                 // Initialized below before the chunk loop (same deferred-init pattern
                 // as `parsedFileStorePath`); this closure only runs from the loop.
                 durableParsedFileStoragePath: durableParsedFileDir,
+                // CFG/PDG opt-in (#2081 M1) — baked into each worker's workerData so the
+                // worker builds + attaches cfgSideChannel. Off by default.
+                pdg: options?.pdg === true,
+                pdgMaxFunctionLines: options?.pdgMaxFunctionLines,
                 // Fan each chunk across the whole pool (#worker-idle): without this a
                 // chunk smaller than the 8 MB sub-batch cap became a single job on a
                 // single worker. Honors an explicit `subBatchMaxBytes` / env override.
@@ -566,7 +571,19 @@ export async function runChunkedParseAndResolve(graph, scannedFiles, allPaths, t
                     filePath: f.path,
                     contentHash: fileContentHash(f.content),
                 }));
-                chunkHash = computeChunkHash(entries);
+                chunkHash = computeChunkHash(entries, 
+                // Only worker-visible pdg config participates in the key —
+                // pdgMaxEdgesPerFunction is emit-time-only and deliberately
+                // excluded (see PdgCacheKey in parse-cache.ts; #2099 F3). The line
+                // cap is RESOLVED to the worker's default before folding so an
+                // explicit-default run shares the default run's keys (the worker
+                // output is byte-identical either way).
+                options?.pdg === true
+                    ? {
+                        pdg: true,
+                        maxFunctionLines: options?.pdgMaxFunctionLines ?? DEFAULT_PDG_MAX_FUNCTION_LINES,
+                    }
+                    : false);
             }
             const cachedRaw = chunkHash && parseCache ? await loadParseCacheChunk(parseCache, chunkHash) : undefined;
             // Track every chunk hash we touched so the orchestrator can

package/dist/core/ingestion/pipeline.d.ts

@@ -25,6 +25,29 @@ export interface PipelineOptions {
      * to retain those nodes under `skipGraphPhases`.
      */
     skipGraphPhases?: boolean;
+    /**
+     * Build the control-flow-graph / PDG substrate (#2081 M1, opt-in via `--pdg`).
+     * Off by default: workers skip all CFG work and emit no `cfgSideChannel`, and
+     * scope-resolution emits no BasicBlock nodes or CFG edges — so the default
+     * graph is byte-identical to a pre-#2081 run. Folded into the parse-cache key
+     * so a pdg-off warm cache is not reused on a `--pdg` run.
+     */
+    pdg?: boolean;
+    /**
+     * Per-function source-line cap for worker-side CFG construction.
+     * `undefined` ⇒ the worker applies `DEFAULT_PDG_MAX_FUNCTION_LINES`; `0` ⇒ no
+     * cap (unlimited). Bounds the cost of a pathological mega-function; over-cap
+     * functions are skipped (no CFG emitted for them). No CLI flag in M1 —
+     * programmatic / server analyze-worker path only.
+     */
+    pdgMaxFunctionLines?: number;
+    /**
+     * Per-function CFG edge cap for the scope-resolution emit step.
+     * `undefined` ⇒ `DEFAULT_MAX_CFG_EDGES_PER_FUNCTION`; `0` ⇒ no cap (unlimited).
+     * Over-cap functions stop at the cap and log a structured drop warning (no
+     * silent truncation). No CLI flag in M1 — programmatic / server path only.
+     */
+    pdgMaxEdgesPerFunction?: number;
     /**
      * Request parsing with the worker pool disabled. The sequential parser was
      * removed — the worker pool is the sole parse path — so setting this now