Add recipes to the depgraph command.

recipe_engine/depgraph.py

Index: recipe_engine/depgraph.py
diff --git a/recipe_engine/depgraph.py b/recipe_engine/depgraph.py
index e12ac38c37b5353bf2d227c8c2a1dd43e47d47f8..0aa5195ab7a7bb068394b7761b9291c34e0e0d85 100644
--- a/recipe_engine/depgraph.py
+++ b/recipe_engine/depgraph.py
@@ -20,29 +20,64 @@ _GRAPH_FOOTER = """}
 """
 
 
-def main(universe, ignore_packages, stdout):
-  packages = {}
+def main(universe, own_package, ignore_packages, stdout):
+  modulify = lambda dep: 'module_' + clean(dep)
+  clean = lambda x: x.replace('.', '_dot_').replace('/', '_slash_').replace(
+      ':', '_colon_').replace('-', '_dash')
+
   module_to_package = {}
   edges = []
   for package, module_name in universe.loop_over_recipe_modules():
     mod = universe.load(package, module_name)
 
     for dep in mod.LOADED_DEPS:
-      edges.append((mod.NAME, dep))
-    packages.setdefault(package.name, []).append(mod.NAME)
+      edges.append(((True, mod.NAME), (True, dep)))

[Paweł Hajdan Jr., 2016/03/30 09:21:58]

It's not clear what's the meaning of True here - could you add a comment what
each element of tuple means?

     module_to_package[mod.NAME] = package.name
 
+  recipe_to_package = {}
+  universe_view = loader.UniverseView(universe, own_package)
+  for path, recipe_name in universe_view.loop_over_recipes():
+    if 'example' in recipe_name:

[Paweł Hajdan Jr., 2016/03/30 09:21:57]

These are hacks. Can we just add command-line arguments allowing filtering
recipe names out?

+      continue
+
+    if 'clank' not in recipe_name:
+      continue
+    recipe = universe_view.load_recipe(recipe_name)
+    for dep in recipe.LOADED_DEPS:
+      edges.append(((False, recipe_name), (True, dep)))

[Paweł Hajdan Jr., 2016/03/30 09:21:57]

IMO recipes in the graph should be optional. I'd prefer the default to not
include recipes, but I don't have a strong opinion about that.

What I care about, is to add a command-line switch to allow including or
excluding recipes here.

+    recipe_to_package[recipe_name] = package
+
   print(_GRAPH_HEADER, file=stdout)
   for edge in edges:
-    if (module_to_package[edge[0]] in ignore_packages or
-        module_to_package[edge[1]] in ignore_packages):
+    (first_mod, first_name), (second_mod, second_name) = edge
+
+    if first_mod:
+      if module_to_package[first_name] in ignore_packages:
+        continue
+      first_name = modulify(first_name)

[Paweł Hajdan Jr., 2016/03/30 09:21:58]

Why do we need calls to modulify?

That's probably to avoid collisions, but then why don't we rather prefix
recipes? Most of the dependencies are actually between modules, and recipes only
have edges going out.

By the way, a possible idea may be for recipes to have a different shape on the
graph, and use a different display label than we use for linking graph nodes.

+    else:
+      if recipe_to_package[first_name] in ignore_packages:
+        continue
+
+    if module_to_package[second_name] in ignore_packages:
       continue
-    print('  %s -> %s' % (edge[0], edge[1]), file=stdout)
+    second_name = modulify(second_name)
+
+    print('  %s -> %s' % (clean(first_name), second_name), file=stdout)
+
+  packages = {}
+  for module, package in module_to_package.iteritems():
+    packages.setdefault(package, []).append(module)
   for package, modules in packages.iteritems():
     if package in ignore_packages:
       continue
     # The "cluster_" prefix has magic meaning for graphviz and makes it
     # draw a box around the subgraph.
     print('  subgraph "cluster_%s" { label="%s"; %s; }' % (
-              package, package, '; '.join(modules)), file=stdout)
+        package, package, '; '.join(
+            modulify(x) for x in modules)), file=stdout)
+
+  recipe_names = [clean(name) for name in recipe_to_package.keys()]
+  print('  subgraph "cluster_recipes" { label="recipes"; %s; }' % (
+      '; '.join(recipe_names)), file=stdout)
   print(_GRAPH_FOOTER, file=stdout)