Merge pull request #17483 from smowton/smowton/feature/csharp-dataflow-fewer-nodes-including-virtual-dispatch

C#: Restrict dataflow node creation to source and source-referenced entities [virtual-dispatch-inclusive variant]

csharp/ql/lib/change-notes/2024-09-19-reduced-dataflow-nodes.md

@@ -0,0 +1,4 @@
+---
+category: minorAnalysis
+---
+* `DataFlow::Node` instances are no longer created for library methods and fields that are not callable (either statically or dynamically) or otherwise referred to from source code. This may affect third-party queries that use these nodes to identify library methods or fields that are present in DLL files where those methods or fields are unreferenced. If this presents a problem, consider using `Callable` and other non-dataflow classes to identify such library entities.

csharp/ql/lib/semmle/code/csharp/dataflow/internal/DataFlowPrivate.qll

@@ -995,6 +995,52 @@ private class InstanceCallable extends Callable {
   Location getARelevantLocation() { result = l }
 }
 
+/**
+ * A callable which is either itself defined in source or which is the target
+ * of some call in source, and therefore ought to have dataflow nodes created.
+ *
+ * Note that for library methods these are always unbound declarations, since
+ * generic instantiations never have dataflow nodes constructed.
+ */
+private class CallableUsedInSource extends Callable {
+  CallableUsedInSource() {
+    // Should generate nodes even for abstract methods declared in source
+    this.fromSource()
+    or
+    // Should generate nodes even for synthetic methods derived from source
+    this.hasBody()
+    or
+    exists(Callable target |
+      exists(Call c |
+        // Note that getADynamicTarget does not always include getTarget.
+        target = c.getTarget()
+        or
+        // Note that getARuntimeTarget cannot be used here, because the
+        // DelegateLikeCall case depends on lambda-flow, which in turn
+        // uses the dataflow library; hence this would introduce recursion
+        // into the definition of data-flow nodes.
+        exists(DispatchCall dc | c = dc.getCall() | target = dc.getADynamicTarget())
+      )
+      or
+      target = any(CallableAccess ca).getTarget()
+    |
+      this = target.getUnboundDeclaration()
+    )
+  }
+}
+
+/**
+ * A field or property which is either itself defined in source or which is the target
+ * of some access in source, and therefore ought to have dataflow nodes created.
+ */
+private class FieldOrPropertyUsedInSource extends FieldOrProperty {
+  FieldOrPropertyUsedInSource() {
+    this.fromSource()
+    or
+    this.getAnAccess().fromSource()
+  }
+}
+
 /** A collection of cached types and predicates to be evaluated in the same stage. */
 cached
 private module Cached {
@@ -1018,8 +1064,13 @@ private module Cached {
     TAssignableDefinitionNode(AssignableDefinition def, ControlFlow::Node cfn) {
       cfn = def.getExpr().getAControlFlowNode()
     } or
-    TExplicitParameterNode(Parameter p, DataFlowCallable c) { p = c.asCallable(_).getAParameter() } or
-    TInstanceParameterNode(InstanceCallable c, Location l) { l = c.getARelevantLocation() } or
+    TExplicitParameterNode(Parameter p, DataFlowCallable c) {
+      p = c.asCallable(_).(CallableUsedInSource).getAParameter()
+    } or
+    TInstanceParameterNode(InstanceCallable c, Location l) {
+      c instanceof CallableUsedInSource and
+      l = c.getARelevantLocation()
+    } or
     TDelegateSelfReferenceNode(Callable c) { lambdaCreationExpr(_, c) } or
     TLocalFunctionCreationNode(ControlFlow::Nodes::ElementNode cfn, Boolean isPostUpdate) {
       cfn.getAstNode() instanceof LocalFunctionStmt
@@ -1055,11 +1106,13 @@ private module Cached {
       or
       lambdaCallExpr(_, cfn)
     } or
-    TFlowSummaryNode(FlowSummaryImpl::Private::SummaryNode sn) or
+    TFlowSummaryNode(FlowSummaryImpl::Private::SummaryNode sn) {
+      sn.getSummarizedCallable() instanceof CallableUsedInSource
+    } or
     TParamsArgumentNode(ControlFlow::Node callCfn) {
       callCfn = any(Call c | isParamsArg(c, _, _)).getAControlFlowNode()
     } or
-    TFlowInsensitiveFieldNode(FieldOrProperty f) { f.isFieldLike() } or
+    TFlowInsensitiveFieldNode(FieldOrPropertyUsedInSource f) { f.isFieldLike() } or
     TFlowInsensitiveCapturedVariableNode(LocalScopeVariable v) { v.isCaptured() } or
     TInstanceParameterAccessNode(ControlFlow::Node cfn, Boolean isPostUpdate) {
       cfn = getAPrimaryConstructorParameterCfn(_)