@codeyam/codeyam-cli 0.1.0-staging.6e699e5 → 0.1.0-staging.7481014

package/analyzer-template/packages/ai/src/lib/astScopes/methodSemantics.ts

@@ -329,11 +329,17 @@ export class ArrayReduceSemantics implements MethodSemantics {
 /**
  * Array concat method semantics
  *
- * concat() returns a new array combining the original array and additional items
+ * concat() returns a new array/string combining the original and additional items
+ *
+ * NOTE: We intentionally do NOT type the source as 'array' here because
+ * .concat() exists on both arrays AND strings. Typing the source prematurely
+ * causes type contamination when the source is equivalent to other paths
+ * via ?? chains. Type inference happens later in checkIfKnownType() based
+ * on accumulated evidence from all methods called on a path.
  */
 export class ArrayConcatSemantics implements MethodSemantics {
   getReturnType(): string {
-    return 'array';
+    return 'unknown'; // Could be array or string depending on source
   }
 
   addEquivalences(
@@ -341,23 +347,24 @@ export class ArrayConcatSemantics implements MethodSemantics {
     sourcePath: StructuredPath,
     context: AnalysisContext,
   ): void {
-    // Mark source as array
-    context.addType(sourcePath, 'array');
+    // NOTE: We use 'unknown' instead of 'array' for the source because .concat()
+    // exists on both arrays AND strings. Using 'array' causes type contamination
+    // when the source is equivalent to other paths via ?? chains.
+    context.addType(sourcePath, 'unknown');
 
-    // Any arguments are also added to the array
+    // Any arguments are also added to the result
     const functionCallSegment = methodCallPath.getLastFunctionCallSegment();
 
     if (functionCallSegment) {
       // Each argument could be an array or single element
       functionCallSegment.args.forEach((argPath) => {
         // No need to guess the type - we'll discover it through normal processing
-        // But we can indicate that these elements will be in the result array
+        // But we can indicate that these elements will be in the result
         context.addEquivalence(methodCallPath, argPath);
       });
     }
 
-    // The original array elements are also in the result
-    context.addType(methodCallPath, 'array');
+    // The original elements are also in the result
     context.addEquivalence(methodCallPath, sourcePath.withElement('*'));
   }
 
@@ -369,11 +376,17 @@ export class ArrayConcatSemantics implements MethodSemantics {
 /**
  * Array slice method semantics
  *
- * slice() returns a shallow copy of a portion of an array
+ * slice() returns a shallow copy of a portion of an array (or string)
+ *
+ * NOTE: We intentionally do NOT type the source as 'array' here because
+ * .slice() exists on both arrays AND strings. Typing the source prematurely
+ * causes type contamination when the source is equivalent to other paths
+ * via ?? chains. Type inference happens later in checkIfKnownType() based
+ * on accumulated evidence from all methods called on a path.
  */
 export class ArraySliceSemantics implements MethodSemantics {
   getReturnType(): string {
-    return 'array';
+    return 'unknown'; // Could be array or string depending on source
   }
 
   addEquivalences(
@@ -381,12 +394,15 @@ export class ArraySliceSemantics implements MethodSemantics {
     sourcePath: StructuredPath,
     context: AnalysisContext,
   ): void {
-    // Mark source as array
-    context.addType(sourcePath, 'array');
+    // NOTE: We use 'unknown' instead of 'array' for the source because .slice()
+    // exists on both arrays AND strings. Using 'array' causes type contamination
+    // when the source is equivalent to other paths via ?? chains.
+    // Type inference happens later in checkIfKnownType() based on accumulated evidence.
+    context.addType(sourcePath, 'unknown');
 
-    // The elements in the result are the same type as the source array
+    // Still track the return value type for data structure tracking
     const methodCallPathWithReturnValues = methodCallPath.withReturnValues();
-    context.addType(methodCallPathWithReturnValues, 'array');
+    context.addType(methodCallPathWithReturnValues, 'unknown');
   }
 
   isComplete(): boolean {
@@ -608,6 +624,12 @@ export class ArrayFlatMapSemantics implements MethodSemantics {
  * Array at method semantics
  *
  * at() returns the element at the given index, supporting negative indices
+ *
+ * NOTE: We intentionally do NOT type the source as 'array' here because
+ * .at() exists on both arrays AND strings. Typing the source prematurely
+ * causes type contamination when the source is equivalent to other paths
+ * via ?? chains. Type inference happens later in checkIfKnownType() based
+ * on accumulated evidence from all methods called on a path.
  */
 export class ArrayAtSemantics implements MethodSemantics {
   getReturnType(): string {
@@ -619,10 +641,12 @@ export class ArrayAtSemantics implements MethodSemantics {
     sourcePath: StructuredPath,
     context: AnalysisContext,
   ): void {
-    // Mark source as array
-    context.addType(sourcePath, 'array');
+    // NOTE: We use 'unknown' instead of 'array' for the source because .at()
+    // exists on both arrays AND strings. Using 'array' causes type contamination
+    // when the source is equivalent to other paths via ?? chains.
+    context.addType(sourcePath, 'unknown');
 
-    // The return value is equivalent to an array element
+    // The return value is equivalent to an element
     context.addEquivalence(
       methodCallPath.withReturnValues(),
       sourcePath.withElement('*'),
@@ -871,6 +895,132 @@ export class UseStateSemantics implements MethodSemantics {
   }
 }
 
+// ─── Set/Map method semantics ──────────────────────────────────────────
+//
+// Set/Map membership operations (.has, .delete, .add, .get, .set, .clear)
+// don't create data flow equivalencies — they're membership checks or
+// mutations, not transformations that propagate schema structure.
+// Without these semantics, code like `filterRatings.has(articleId)` creates
+// spurious equivalencies (332 entries in the Margo LibraryPage case).
+
+/**
+ * Set.has() / Map.has() — membership check, returns boolean
+ */
+export class CollectionHasSemantics implements MethodSemantics {
+  getReturnType(): string {
+    return 'boolean';
+  }
+
+  addEquivalences(
+    methodCallPath: StructuredPath,
+    _sourcePath: StructuredPath,
+    context: AnalysisContext,
+  ): void {
+    context.addType(methodCallPath, 'boolean');
+  }
+
+  isComplete(): boolean {
+    return true;
+  }
+}
+
+/**
+ * Set.delete() / Map.delete() — removal, returns boolean
+ */
+export class CollectionDeleteSemantics implements MethodSemantics {
+  getReturnType(): string {
+    return 'boolean';
+  }
+
+  addEquivalences(
+    methodCallPath: StructuredPath,
+    _sourcePath: StructuredPath,
+    context: AnalysisContext,
+  ): void {
+    context.addType(methodCallPath, 'boolean');
+  }
+
+  isComplete(): boolean {
+    return true;
+  }
+}
+
+/**
+ * Set.add() — adds an element, returns the Set (for chaining)
+ */
+export class SetAddSemantics implements MethodSemantics {
+  getReturnType(): string {
+    return 'object'; // Returns the Set itself
+  }
+
+  addEquivalences(
+    _methodCallPath: StructuredPath,
+    _sourcePath: StructuredPath,
+    _context: AnalysisContext,
+  ): void {
+    // No equivalencies — add() doesn't create data flow
+  }
+
+  isComplete(): boolean {
+    return true;
+  }
+}
+
+/**
+ * Set.clear() / Map.clear() — removes all elements, returns void
+ */
+export class CollectionClearSemantics implements MethodSemantics {
+  getReturnType(): string {
+    return 'void';
+  }
+
+  addEquivalences(): void {
+    // No equivalencies
+  }
+
+  isComplete(): boolean {
+    return true;
+  }
+}
+
+/**
+ * Map.get() — retrieves a value by key, returns unknown (value type)
+ */
+export class MapGetSemantics implements MethodSemantics {
+  getReturnType(): string {
+    return 'unknown';
+  }
+
+  addEquivalences(
+    _methodCallPath: StructuredPath,
+    _sourcePath: StructuredPath,
+    _context: AnalysisContext,
+  ): void {
+    // No equivalencies — the return value type depends on the map contents
+  }
+
+  isComplete(): boolean {
+    return true;
+  }
+}
+
+/**
+ * Map.set() — sets a key-value pair, returns the Map (for chaining)
+ */
+export class MapSetSemantics implements MethodSemantics {
+  getReturnType(): string {
+    return 'object'; // Returns the Map itself
+  }
+
+  addEquivalences(): void {
+    // No equivalencies
+  }
+
+  isComplete(): boolean {
+    return true;
+  }
+}
+
 /**
  * Create and populate the registry with method semantics
  */
@@ -906,6 +1056,15 @@ export function createMethodRegistry(): MethodSemanticsRegistry {
   registry.register('useState', new UseStateSemantics(), 'React');
   registry.register('useMemo', new UseMemoSemantics(), 'React');
 
+  // Register Set/Map collection methods.
+  // Only .has() and .delete() are safe to register by method name — they always
+  // return boolean regardless of the collection type (Set, Map, WeakSet, WeakMap,
+  // FormData, URLSearchParams, etc.).
+  // .add(), .clear(), .get(), .set() are too ambiguous across different types
+  // and can't be registered until the lookup code supports class-aware resolution.
+  registry.register('has', new CollectionHasSemantics());
+  registry.register('delete', new CollectionDeleteSemantics());
+
   return registry;
 }
 

package/analyzer-template/packages/ai/src/lib/astScopes/nodeToSource.ts

@@ -54,6 +54,25 @@ export function nodeToSource(
   //   return serializeBindingElement(node, sourceFile, options);
   // }
 
+  // Object/array/function literals should be marked as literal paths.
+  // This prevents them from creating sourceEquivalencies when used as
+  // function call arguments (e.g., setUndoEntry({ label: '...', undo: ... })).
+  // Without this, the full source text of the literal (300-500 chars for complex
+  // objects) gets embedded in path strings, creating expensive-to-parse
+  // equivalency entries that don't contribute useful schema information.
+  if (ts.isObjectLiteralExpression(node)) {
+    return StructuredPath.createLiteralPath(node.getText(sourceFile), 'object');
+  }
+  if (ts.isArrayLiteralExpression(node)) {
+    return StructuredPath.createLiteralPath(node.getText(sourceFile), 'array');
+  }
+  if (ts.isArrowFunction(node) || ts.isFunctionExpression(node)) {
+    return StructuredPath.createLiteralPath(
+      node.getText(sourceFile),
+      'function',
+    );
+  }
+
   // Fallback for unsupported node types
   return StructuredPath.fromBase(node.getText(sourceFile));
 }

package/analyzer-template/packages/ai/src/lib/astScopes/paths.ts

@@ -406,11 +406,18 @@ export class StructuredPath {
           } else {
             // Convert each argument to its string representation
             const argsStr = funcSegment.args
-              .map((arg) =>
-                arg.toStringWithSegments(arg.base, arg.segments, {
+              .map((arg) => {
+                // For literal arguments (object/array/function literals),
+                // use a compact placeholder instead of the full source text.
+                // This prevents 300-500 char inline objects from being embedded
+                // in call signature paths, which creates expensive equivalencies.
+                if (arg.isLiteral()) {
+                  return `{...}`;
+                }
+                return arg.toStringWithSegments(arg.base, arg.segments, {
                   withReturnValues: doRenderNestedFunctionCallReturnValue,
-                }),
-              )
+                });
+              })
               .join(', ');
             result += `(${argsStr})`;
           }

package/analyzer-template/packages/ai/src/lib/astScopes/patterns/forInStatementHandler.ts

@@ -22,11 +22,8 @@ export class ForInStatementHandler implements PatternHandler {
     // Process the expression being iterated over first (e.g., object)
     context.processExpression(node.expression);
 
-    // Get the path of the object being iterated
-    const expressionPath = context.getStructuralPath(node.expression);
-
-    // If the initializer is a variable declaration with an identifier, establish equivalence
-    // to indicate it's iterating over the keys of the expression
+    // If the initializer is a variable declaration with an identifier,
+    // type it as 'string' since for-in keys are always strings
     if (ts.isVariableDeclarationList(node.initializer)) {
       const declarations = node.initializer.declarations;
       if (
@@ -36,25 +33,21 @@ export class ForInStatementHandler implements PatternHandler {
       ) {
         const variableName = StructuredPath.fromBase(declarations[0].name.text);
 
-        // Add type for the key variable (string for object keys)
+        // Type the key variable as 'string' — for-in keys are always strings.
+        // Note: We intentionally do NOT equivalence the key variable to
+        // expressionPath[*key*]. The key variable is the KEY (a string),
+        // while [*key*] in the schema represents the VALUE at a dynamic key.
+        // Creating that equivalence would cause the 'string' type to propagate
+        // to the value path. The value access (e.g., obj[key]) is handled
+        // correctly by the element access handler in processExpression.
         context.addType(variableName, 'string');
-
-        if (expressionPath) {
-          // For for...in loops, the variable is equivalent to the keys of the object
-          // We represent this with a *key* suffix to indicate it's a key, not a value
-          context.addEquivalence(variableName, expressionPath.withKey('key'));
-        }
       }
     } else if (ts.isIdentifier(node.initializer)) {
       // If the initializer is directly an identifier (rare but possible)
       const variableName = StructuredPath.fromBase(node.initializer.text);
 
-      // Add type for the key variable
+      // See comment above — type as string but don't equivalence to value path
       context.addType(variableName, 'string');
-
-      if (expressionPath) {
-        context.addEquivalence(variableName, expressionPath.withKey('key'));
-      }
     } else {
       context.processExpression(node.initializer);
     }