gyp: fix O(n^2) in dependency calculations.

pylib/gyp/input.py

 # Copyright (c) 2012 Google Inc. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 from compiler.ast import Const
 from compiler.ast import Dict
 from compiler.ast import Discard
 from compiler.ast import List
 from compiler.ast import Module
 from compiler.ast import Node
@@ skipping 1462 matching lines @@
 
   def __repr__(self):
     return '<DependencyGraphNode: %r>' % self.ref
 
   def FlattenToList(self):
     # flat_list is the sorted list of dependencies - actually, the list items
     # are the "ref" attributes of DependencyGraphNodes.  Every target will
     # appear in flat_list after all of its dependencies, and before all of its
     # dependents.
     flat_list = []
+    flat_set = set()

[scottmg, 2014/04/11 18:28:26]

comment here to the effect that _set is for optimization and the two must be
coherent.

[Daniel Bratell, 2014/04/14 12:18:27]

Done.

 
     # in_degree_zeros is the list of DependencyGraphNodes that have no
     # dependencies not in flat_list.  Initially, it is a copy of the children
     # of this node, because when the graph was built, nodes with no
     # dependencies were made implicit dependents of the root node.
     in_degree_zeros = set(self.dependents[:])
 
     while in_degree_zeros:
       # Nodes in in_degree_zeros have no dependencies not in flat_list, so they
       # can be appended to flat_list.  Take these nodes out of in_degree_zeros
       # as work progresses, so that the next node to process from the list can
       # always be accessed at a consistent position.
       node = in_degree_zeros.pop()
       flat_list.append(node.ref)
+      flat_set.add(node.ref)
 
       # Look at dependents of the node just added to flat_list.  Some of them
       # may now belong in in_degree_zeros.
       for node_dependent in node.dependents:
         is_in_degree_zero = True
+        # TODO: We want to check through the
+        # node_dependent.dependencies list but if it's long and we
+        # always start at the beginning, then we get O(n^2) behaviour.
         for node_dependent_dependency in node_dependent.dependencies:
-          if not node_dependent_dependency.ref in flat_list:
+          if not node_dependent_dependency.ref in flat_set:
             # The dependent one or more dependencies not in flat_list.  There
             # will be more chances to add it to flat_list when examining
             # it again as a dependent of those other dependencies, provided
             # that there are no cycles.
             is_in_degree_zero = False
             break
 
         if is_in_degree_zero:
           # All of the dependent's dependencies are already in flat_list.  Add
           # it to in_degree_zeros where it will be processed in a future
@@ skipping 1301 matching lines @@
     ValidateRunAsInTarget(target, target_dict, build_file)
     ValidateActionsInTarget(target, target_dict, build_file)
 
   # Generators might not expect ints.  Turn them into strs.
   TurnIntIntoStrInDict(data)
 
   # TODO(mark): Return |data| for now because the generator needs a list of
   # build files that came in.  In the future, maybe it should just accept
   # a list, and not the whole data dict.
   return [flat_list, targets, data]