Merge pull request #9700 from jketema/resolve-global-variable

C++: Ensure only one `Variable` exists for every global variable

cpp/ql/lib/change-notes/2022-06-24-unique-variable.md

@@ -0,0 +1,4 @@
+---
+category: fix
+---
+* Under certain circumstances a variable declaration that is not also a definition could be associated with a `Variable` that did not have the definition as a `VariableDeclarationEntry`. This is now fixed, and a unique `Variable` will exist that has both the declaration and the definition as a `VariableDeclarationEntry`.

cpp/ql/lib/semmle/code/cpp/Element.qll

@@ -6,6 +6,7 @@
 import semmle.code.cpp.Location
 private import semmle.code.cpp.Enclosing
 private import semmle.code.cpp.internal.ResolveClass
+private import semmle.code.cpp.internal.ResolveGlobalVariable
 
 /**
  * Get the `Element` that represents this `@element`.
@@ -28,9 +29,12 @@ Element mkElement(@element e) { unresolveElement(result) = e }
 pragma[inline]
 @element unresolveElement(Element e) {
   not result instanceof @usertype and
+  not result instanceof @variable and
   result = e
   or
   e = resolveClass(result)
+  or
+  e = resolveGlobalVariable(result)
 }
 
 /**

cpp/ql/lib/semmle/code/cpp/Variable.qll

@@ -6,6 +6,7 @@ import semmle.code.cpp.Element
 import semmle.code.cpp.exprs.Access
 import semmle.code.cpp.Initializer
 private import semmle.code.cpp.internal.ResolveClass
+private import semmle.code.cpp.internal.ResolveGlobalVariable
 
 /**
  * A C/C++ variable. For example, in the following code there are four
@@ -32,6 +33,8 @@ private import semmle.code.cpp.internal.ResolveClass
  * can have multiple declarations.
  */
 class Variable extends Declaration, @variable {
+  Variable() { isVariable(underlyingElement(this)) }
+
   override string getAPrimaryQlClass() { result = "Variable" }
 
   /** Gets the initializer of this variable, if any. */

cpp/ql/lib/semmle/code/cpp/internal/ResolveGlobalVariable.qll

@@ -0,0 +1,60 @@
+private predicate hasDefinition(@globalvariable g) {
+  exists(@var_decl vd | var_decls(vd, g, _, _, _) | var_def(vd))
+}
+
+pragma[noinline]
+private predicate onlyOneCompleteGlobalVariableExistsWithMangledName(@mangledname name) {
+  strictcount(@globalvariable g | hasDefinition(g) and mangled_name(g, name)) = 1
+}
+
+/** Holds if `g` is a unique global variable with a definition named `name`. */
+pragma[noinline]
+private predicate isGlobalWithMangledNameAndWithDefinition(@mangledname name, @globalvariable g) {
+  hasDefinition(g) and
+  mangled_name(g, name) and
+  onlyOneCompleteGlobalVariableExistsWithMangledName(name)
+}
+
+/** Holds if `g` is a global variable without a definition named `name`. */
+pragma[noinline]
+private predicate isGlobalWithMangledNameAndWithoutDefinition(@mangledname name, @globalvariable g) {
+  not hasDefinition(g) and
+  mangled_name(g, name)
+}
+
+/**
+ * Holds if `incomplete` is a global variable without a definition, and there exists
+ * a unique global variable `complete` with the same name that does have a definition.
+ */
+private predicate hasTwinWithDefinition(@globalvariable incomplete, @globalvariable complete) {
+  exists(@mangledname name |
+    not variable_instantiation(incomplete, complete) and
+    isGlobalWithMangledNameAndWithoutDefinition(name, incomplete) and
+    isGlobalWithMangledNameAndWithDefinition(name, complete)
+  )
+}
+
+import Cached
+
+cached
+private module Cached {
+  /**
+   * If `v` is a global variable without a definition, and there exists a unique
+   * global variable with the same name that does have a definition, then the
+   * result is that unique global variable. Otherwise, the result is `v`.
+   */
+  cached
+  @variable resolveGlobalVariable(@variable v) {
+    hasTwinWithDefinition(v, result)
+    or
+    not hasTwinWithDefinition(v, _) and
+    result = v
+  }
+
+  cached
+  predicate isVariable(@variable v) {
+    not v instanceof @globalvariable
+    or
+    v = resolveGlobalVariable(_)
+  }
+}