You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

100 lines
2.8 KiB

#!/usr/bin/env python
# Copyright (c) 2011 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.
"""Using the JSON dumped by the dump-dependency-json generator,
generate input suitable for graphviz to render a dependency graph of
targets."""
import collections
import json
import sys
def ParseTarget(target):
target, _, suffix = target.partition('#')
filename, _, target = target.partition(':')
return filename, target, suffix
def LoadEdges(filename, targets):
"""Load the edges map from the dump file, and filter it to only
show targets in |targets| and their depedendents."""
file = open('dump.json')
edges = json.load(file)
file.close()
# Copy out only the edges we're interested in from the full edge list.
target_edges = {}
to_visit = targets[:]
while to_visit:
src = to_visit.pop()
if src in target_edges:
continue
target_edges[src] = edges[src]
to_visit.extend(edges[src])
return target_edges
def WriteGraph(edges):
"""Print a graphviz graph to stdout.
|edges| is a map of target to a list of other targets it depends on."""
# Bucket targets by file.
files = collections.defaultdict(list)
for src, dst in edges.items():
build_file, target_name, toolset = ParseTarget(src)
files[build_file].append(src)
print 'digraph D {'
print ' fontsize=8' # Used by subgraphs.
print ' node [fontsize=8]'
# Output nodes by file. We must first write out each node within
# its file grouping before writing out any edges that may refer
# to those nodes.
for filename, targets in files.items():
if len(targets) == 1:
# If there's only one node for this file, simplify
# the display by making it a box without an internal node.
target = targets[0]
build_file, target_name, toolset = ParseTarget(target)
print ' "%s" [shape=box, label="%s\\n%s"]' % (target, filename,
target_name)
else:
# Group multiple nodes together in a subgraph.
print ' subgraph "cluster_%s" {' % filename
print ' label = "%s"' % filename
for target in targets:
build_file, target_name, toolset = ParseTarget(target)
print ' "%s" [label="%s"]' % (target, target_name)
print ' }'
# Now that we've placed all the nodes within subgraphs, output all
# the edges between nodes.
for src, dsts in edges.items():
for dst in dsts:
print ' "%s" -> "%s"' % (src, dst)
print '}'
def main():
if len(sys.argv) < 2:
print >>sys.stderr, __doc__
print >>sys.stderr
print >>sys.stderr, 'usage: %s target1 target2...' % (sys.argv[0])
return 1
edges = LoadEdges('dump.json', sys.argv[1:])
WriteGraph(edges)
return 0
if __name__ == '__main__':
sys.exit(main())