DataFlow:Run overlay-informed if not diff-informed

To ensure good performance, always run data flow overlay-informed unless the configuration has opted in to being diff-informed. This change affects only databases with an overlay and therefore has no immediate production consequences.
github · jbj · Jun 16, 2025 · Jun 16, 2025 · Jun 20, 2025 · Jan 15, 2025
commit c1214db5b3c482975035d31183ddb503af1cb04c
diff --git a/java/ql/lib/semmle/code/java/dataflow/internal/DataFlowImplSpecific.qll b/java/ql/lib/semmle/code/java/dataflow/internal/DataFlowImplSpecific.qll
@@ -6,6 +6,7 @@ module;
 
 private import semmle.code.Location
 private import codeql.dataflow.DataFlow
+private import semmle.code.java.Overlay
 
 module Private {
   import DataFlowPrivate
@@ -29,4 +30,6 @@ module JavaDataFlow implements InputSig<Location> {
   predicate mayBenefitFromCallContext = Private::mayBenefitFromCallContext/1;
 
   predicate viableImplInCallContext = Private::viableImplInCallContext/2;
+
+  predicate isEvaluatingInOverlay = isOverlay/0;
 }
diff --git a/shared/dataflow/codeql/dataflow/DataFlow.qll b/shared/dataflow/codeql/dataflow/DataFlow.qll
@@ -349,6 +349,18 @@ signature module InputSig<LocationSig Location> {
 
   /** Holds if `fieldFlowBranchLimit` should be ignored for flow going into/out of `c`. */
   default predicate ignoreFieldFlowBranchLimit(DataFlowCallable c) { none() }
+
+  /**
+   * Holds if the evaluator is currently evaluating with an overlay. The
+   * implementation of this predicate needs to be `overlay[local]`. For a
+   * language with no overlay support, `none()` is a valid implementation.
+   *
+   * When called from a local predicate, this predicate holds if we are in the
+   * overlay-only local evaluation. When called from a global pre
8000
dicate, this
+   * predicate holds if we are evaluating globally with overlay and base both
+   * visible.
+   */
-   */
+   */
+  overlay[local]
-   */
+   */
+  overlay[local]
+  default predicate isEvaluatingInOverlay() { none() }
 }
 
 module Configs<LocationSig Location, InputSig<Location> Lang> {

diff --git a/shared/dataflow/codeql/dataflow/internal/DataFlowImplStage1.qll b/shared/dataflow/codeql/dataflow/internal/DataFlowImplStage1.qll
@@ -4,7 +4,7 @@
  * Provides an implementation of a fast initial pruning of global
  * (interprocedural) data flow reachability (Stage 1).
  */
-overlay[local?]
+overlay[local?] // when this is removed, put `overlay[local?]` on `isOverlayNode`.
 module;
 
 private import codeql.util.Unit
@@ -129,14 +129,38 @@ module MakeImplStage1<LocationSig Location, InputSig<Location> Lang> {
 
       private module AlertFiltering = AlertFilteringImpl<Location>;
 
+      /**
+       * Holds if the given node is visible in overlay-only local evaluation.
+       *
+       * This predicate needs to be `overlay[local?]`, either directly or
+       * through annotations from an outer scope. If `Node` is global for the
+       * language under analysis, then every node is considered an overlay
+       * node, which means there will effectively be no overlay-based
+       * filtering of sources and sinks.
+       */
+      private predicate isOverlayNode(Node node) {
+        isEvaluatingInOverlay() and
+        // Any local node is an overlay node if we are evaluating in overlay mode
+        node = node
+      }
+
+      overlay[global]
       pragma[nomagic]
       private predicate isFilteredSource(Node source) {
         Config::isSource(source, _) and
         if Config::observeDiffInformedIncrementalMode()
         then AlertFiltering::filterByLocation(Config::getASelectedSourceLocation(source))
-        else any()
+        else (
+          // If we are in base-only global evaluation, do not filter out any sources.
+          not isEvaluatingInOverlay()
+          or
+          // If we are in global evaluation with an overlay present, restrict
+          // sources to those visible in the overlay.
+          isOverlayNode(source)
+        )
       }
 
+      overlay[global]
       pragma[nomagic]
       private predicate isFilteredSink(Node sink) {
         (
@@ -145,7 +169,14 @@ module MakeImplStage1<LocationSig Location, InputSig<Location> Lang> {
         ) and
         if Config::observeDiffInformedIncrementalMode()
         then AlertFiltering::filterByLocation(Config::getASelectedSinkLocation(sink))
-        else any()
+        else (
+          // If we are in base-only global evaluation, do not filter out any sinks.
+          not isEvaluatingInOverlay()
+          or
+          // If we are in global evaluation with an overlay present, restrict
+          // sinks to those visible in the overlay.
+          isOverlayNode(sink)
+        )
       }
 
       private predicate hasFilteredSource() { isFilteredSource(_) }