github · yoff · Mar 12, 2021 · Mar 11, 2021 · Mar 12, 2021
@@ -0,0 +1,2 @@
+lgtm,codescanning
+* API graphs now contain nodes for built-in functions and classes. For instance, `API::builtin("open")` is the API graph node corresponding to the built-in `open` function.
@@ -216,6 +216,9 @@ module API {
    */
   Node moduleImport(string m) { result = Impl::MkModuleImport(m) }
 
+  /** Gets a node corresponding to the built-in with the given name, if any. */
+  Node builtin(string n) { result = moduleImport("builtins").getMember(n) }
+
   /**
    * Provides the actual implementation of API graphs, cached for performance.
    *
@@ -300,11 +303,18 @@ module API {
       MkRoot() or
       /** An abstract representative for imports of the module called `name`. */
       MkModuleImport(string name) {
-        imports(_, name)
+        // Ignore the following module name for Python 2, as we alias `__builtin__` to `builtins` elsewhere
+        (name != "__builtin__" or major_version() = 3) and
+        (
+          imports(_, name)
+          or
+          // When we `import foo.bar.baz` we want to create API graph nodes also for the prefixes
+          // `foo` and `foo.bar`:
+          name = any(ImportExpr e | not e.isRelative()).getAnImportedModuleName()
+        )
         or
-        // When we `import foo.bar.baz` we want to create API graph nodes also for the prefixes
-        // `foo` and `foo.bar`:
-        name = any(ImportExpr e | not e.isRelative()).getAnImportedModuleName()
+        // The `builtins` module should always be implicitly available
+        name = "builtins"
       } or
       /** A use of an API member at the node `nd`. */
       MkUse(DataFlow::Node nd) { use(_, _, nd) }
@@ -339,6 +349,24 @@ module API {
       )
     }
 
+    private import semmle.python.types.Builtins as Builtins
+
+    /**
+     * Gets a data flow node that is likely to refer to a built-in with the name `name`.
+     *
+     * Currently this is an over-approximation, and does not account for things like overwriting a
+     * built-in with a different value.
+     */
+    private DataFlow::Node likely_builtin(string name) {
+      result.asCfgNode() =
+        any(NameNode n |
+          n.isGlobal() and
+          n.isLoad() and
+          name = n.getId() and
+          name = any(Builtins::Builtin b).getName()
+        )
+    }
+
     /**
      * Holds if `ref` is a use of a node that should have an incoming edge from `base` labeled
      * `lbl` in the API graph.
@@ -369,6 +397,10 @@ module API {
           ref.asExpr().(ClassExpr).getABase() = superclass.asExpr()
         )
       )
+      or
+      // Built-ins, treated as members of the module `builtins`
+      base = MkModuleImport("builtins") and
+      lbl = Label::member(any(string name | ref = likely_builtin(name)))
     }
 
     /**
@@ -381,6 +413,11 @@ module API {
         imports(ref, name)
       )
       or
+      // Ensure the Python 2 `__builtin__` module gets the name of the Python 3 `builtins` module.
+      major_version() = 2 and
+      nd = MkModuleImport("builtins") and
+      imports(ref, "__builtin__")
+      or
       nd = MkUse(ref)
     }
 

@@ -0,0 +1 @@
+semmle-extractor-options: --lang=2 --max-import-depth=1
@@ -0,0 +1,3 @@
+def python2_style():
+    from __builtin__ import open #$ use=moduleImport("builtins").getMember("open")
+    open("hello.txt") #$ use=moduleImport("builtins").getMember("open").getReturn()
@@ -0,0 +1,30 @@
+import python
+import semmle.python.dataflow.new.DataFlow
+import TestUtilities.InlineExpectationsTest
+import semmle.python.ApiGraphs
+
+class ApiUseTest extends InlineExpectationsTest {
+  ApiUseTest() { this = "ApiUseTest" }
+
+  override string getARelevantTag() { result = "use" }
+
+  private predicate relevant_node(API::Node a, DataFlow::Node n, Location l) {
+    n = a.getAUse() and l = n.getLocation()
+  }
+
+  override predicate hasActualResult(Location location, string element, string tag, string value) {
+    exists(API::Node a, DataFlow::Node n | relevant_node(a, n, location) |
+      tag = "use" and
+      // Only report the longest path on this line:
+      value =
+        max(API::Node a2, Location l2 |
+          relevant_node(a2, _, l2) and
+          l2.getFile() = location.getFile() and
+          l2.getStartLine() = location.getStartLine()
+        |
+          a2.getPath()
+        ) and
+      element = n.toString()
+    )
+  }
+}
@@ -100,3 +100,42 @@ def my_internal_method(self): #$ use=moduleImport("pflask").getMember("views").g
             pass
 
     int_instance = IntMyView() #$ use=moduleImport("pflask").getMember("views").getMember("View").getASubclass().getReturn()
+
+
+# Built-ins
+
+def use_of_builtins():
+    for x in range(5): #$ use=moduleImport("builtins").getMember("range").getReturn()
+        if x < len([]): #$ use=moduleImport("builtins").getMember("len").getReturn()
+            print("Hello") #$ use=moduleImport("builtins").getMember("print").getReturn()
+            raise Exception("Farewell") #$ use=moduleImport("builtins").getMember("Exception").getReturn()
+
+def imported_builtins():
+    import builtins #$ use=moduleImport("builtins")
+    def open(f):
+        return builtins.open(f) #$ MISSING: use=moduleImport("builtins").getMember("open").getReturn()
+
+def redefine_print():
+    def my_print(x):
+        import builtins #$ use=moduleImport("builtins")
+        builtins.print("I'm printing", x) #$ use=moduleImport("builtins").getMember("print").getReturn()
+    print = my_print
+    print("these words")
+
+def local_redefine_range():
+    range = 5
+    return range
+
+def global_redefine_range():
+    global range
+    range = 6
+    return range #$ SPURIOUS: use=moduleImport("builtins").getMember("range")
+
+def obscured_print():
+    p = print #$ use=moduleImport("builtins").getMember("print")
+    p("Can you see me?") #$ use=moduleImport("builtins").getMember("print").getReturn()
+
+def python2_style():
+    # In Python 3, `__builtin__` has no special meaning.
+    from __builtin__ import open #$ use=moduleImport("__builtin__").getMember("open")
+    open("hello.txt") #$ use=moduleImport("__builtin__").getMember("open").getReturn()
@@ -2,5 +2,5 @@
 print(mypkg.foo) #$ use=moduleImport("mypkg").getMember("foo") // 42
 try:
     print(mypkg.bar) #$ use=moduleImport("mypkg").getMember("bar")
-except AttributeError as e:
-    print(e)  # module 'mypkg' has no attribute 'bar'
+except AttributeError as e: #$ use=moduleImport("builtins").getMember("AttributeError")
+    print(e)  #$ use=moduleImport("builtins").getMember("print").getReturn() // module 'mypkg' has no attribute 'bar'
@@ -3,8 +3,8 @@
 print(mypkg.foo) #$ use=moduleImport("mypkg").getMember("foo") // 42
 try:
     print(mypkg.bar) #$ use=moduleImport("mypkg").getMember("bar")
-except AttributeError as e:
-    print(e)  # module 'mypkg' has no attribute 'bar'
+except AttributeError as e: #$ use=moduleImport("builtins").getMember("AttributeError")
+    print(e)  #$ use=moduleImport("builtins").getMember("print").getReturn() // module 'mypkg' has no attribute 'bar'
 
 from mypkg import bar as _bar #$ use=moduleImport("mypkg").getMember("bar")
 print(mypkg.bar) #$ use=moduleImport("mypkg").getMember("bar") // <module 'mypkg.bar' ...
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,2 @@
		lgtm,codescanning
		* API graphs now contain nodes for built-in functions and classes. For instance, `API::builtin("open")` is the API graph node corresponding to the built-in `open` function.
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		semmle-extractor-options: --lang=2 --max-import-depth=1