Python: Remove support for late *args arguments

I found this to cause bad performance, so the implementation of this has
to be thought out more carefully.

Author: RasmusWL

python/ql/lib/semmle/python/dataflow/new/internal/DataFlowDispatch.qll

@@ -64,7 +64,6 @@ newtype TParameterPosition =
     index = any(Parameter p).getPosition() + 1
   } or
   TSynthStarArgsElementParameterPosition(int index) { exists(TStarArgsParameterPosition(index)) } or
-  TSynthLateStarArgsParameterPosition(int index) { exists(TStarArgsParameterPosition(index)) } or
   TDictSplatParameterPosition()
 
 /** A parameter position. */
@@ -90,14 +89,6 @@ class ParameterPosition extends TParameterPosition {
     this = TSynthStarArgsElementParameterPosition(index)
   }
 
-  /**
-   * Holds if this position represents a synthetic `*args` parameter after the real
-   * `*args` parameter. The real `*args` parameter is at the 0-based index `index`.
-   */
-  predicate isSynthLateStarArgsParameterPosition(int index) {
-    this = TSynthLateStarArgsParameterPosition(index)
-  }
-
   /** Holds if this position represents a `**kwargs` parameter. */
   predicate isDictSplat() { this = TDictSplatParameterPosition() }
 
@@ -116,11 +107,6 @@ class ParameterPosition extends TParameterPosition {
       result = "synthetic *args element at (or after) " + index
     )
     or
-    exists(int index |
-      this.isSynthLateStarArgsParameterPosition(index) and
-      result = "synthetic late *args after " + index
-    )
-    or
     this.isDictSplat() and result = "**"
   }
 }
@@ -192,10 +178,6 @@ predicate parameterMatch(ParameterPosition ppos, ArgumentPosition apos) {
     ppos.isSynthStarArgsElement(paramIndex) and apos.isPositional(argIndex)
   )
   or
-  exists(int realStarArgsIndex, int argIndex | argIndex > realStarArgsIndex |
-    ppos.isSynthLateStarArgsParameterPosition(realStarArgsIndex) and apos.isStarArgs(argIndex)
-  )
-  or
   ppos.isDictSplat() and apos.isDictSplat()
 }
 
@@ -329,9 +311,6 @@ abstract class DataFlowFunction extends DataFlowCallable, TFunction {
         or
         ppos.isSynthStarArgsElement(index) and
         result = TSynthStarArgsElementParameterNode(this)
-        or
-        ppos.isSynthLateStarArgsParameterPosition(index) and
-        result = TSynthLateStarArgsParameterNode(this)
       )
     |
       // a `*args` parameter comes after the last positional parameter. We need to take

python/ql/lib/semmle/python/dataflow/new/internal/DataFlowPrivate.qll

@@ -135,49 +135,6 @@ predicate synthStarArgsElementParameterNodeStoreStep(
   )
 }
 
-/**
- * A synthetic node to capture a `*args` argument that is passed to a `*args`
- * parameter, but "too late" in the argument list, so we cannot just do a 1-1 mapping
- * without messing up the indexes; instead we make a list/tuple/set read step to
- * `SynthStarArgsElementParameterNode`.
- *
- * Example. The `*args` arguments starts at index 1, while the `*args` parameter accepts
- * arguments starting at index 0.
- *
- * ```py
- * def func(*args): ...
- * func(1, *args)
- */
-class SynthLateStarArgsParameterNode extends ParameterNodeImpl, TSynthLateStarArgsParameterNode {
-  DataFlowCallable callable;
-
-  SynthLateStarArgsParameterNode() { this = TSynthLateStarArgsParameterNode(callable) }
-
-  override string toString() { result = "SynthLateStarArgsParameterNode" }
-
-  override Scope getScope() { result = callable.getScope() }
-
-  override Location getLocation() { result = callable.getLocation() }
-
-  override Parameter getParameter() { none() }
-}
-
-predicate synthLateStarArgsParameterNodeReadStep(
-  SynthLateStarArgsParameterNode nodeFrom, Content c, ParameterNode nodeTo
-) {
-  (
-    c instanceof ListElementContent
-    or
-    c instanceof TupleElementContent
-    or
-    c instanceof SetElementContent
-  ) and
-  exists(DataFlowCallable callable |
-    nodeFrom = TSynthLateStarArgsParameterNode(callable) and
-    nodeTo = TSynthStarArgsElementParameterNode(callable)
-  )
-}
-
 // =============================================================================
 // **kwargs (DictSplat) related
 // =============================================================================
@@ -843,8 +800,6 @@ predicate readStep(Node nodeFrom, Content c, Node nodeTo) {
   or
   FlowSummaryImpl::Private::Steps::summaryReadStep(nodeFrom, c, nodeTo)
   or
-  synthLateStarArgsParameterNodeReadStep(nodeFrom, c, nodeTo)
-  or
   synthDictSplatParameterNodeReadStep(nodeFrom, c, nodeTo)
 }
 
@@ -1006,8 +961,6 @@ predicate nodeIsHidden(Node n) {
   or
   n instanceof SynthStarArgsElementParameterNode
   or
-  n instanceof SynthLateStarArgsParameterNode
-  or
   n instanceof SynthDictSplatArgumentNode
   or
   n instanceof SynthDictSplatParameterNode

python/ql/lib/semmle/python/dataflow/new/internal/DataFlowPublic.qll

@@ -117,14 +117,6 @@ newtype TNode =
   TSynthStarArgsElementParameterNode(DataFlowCallable callable) {
     exists(ParameterPosition ppos | ppos.isStarArgs(_) | exists(callable.getParameter(ppos)))
   } or
-  /**
-   * A synthetic node to capture a `*args` argument that is passed to a `*args`
-   * parameter, but "too late" in the argument list, so we cannot just do a 1-1 mapping
-   * without messing up the indexes.
-   */
-  TSynthLateStarArgsParameterNode(DataFlowCallable callable) {
-    exists(ParameterPosition ppos | ppos.isStarArgs(_) | exists(callable.getParameter(ppos)))
-  } or
   /** A synthetic node to capture keyword arguments that are passed to a `**kwargs` parameter. */
   TSynthDictSplatArgumentNode(CallNode call) { exists(call.getArgByName(_)) } or
   /** A synthetic node to allow flow to keyword parameters from a `**kwargs` argument. */

python/ql/test/experimental/dataflow/coverage/argumentPassing.py

@@ -213,7 +213,7 @@ def starargs_only(*args):
 def test_only_starargs():
     starargs_only(arg1, arg2, "safe") # $ arg1 arg2 SPURIOUS: bad2,bad3="arg1" bad1,bad3="arg2"
 
-    args = (arg2, "safe") # $ arg2 func=starargs_only SPURIOUS: bad1,bad3="arg2"
+    args = (arg2, "safe") # $ MISSING: arg2
     starargs_only(arg1, *args) # $ arg1 SPURIOUS: bad2,bad3="arg1"
 
     args = (arg1, arg2, "safe") # $ arg1 arg2 func=starargs_only