Fix problem with non void return types in extension API IDL.

tools/json_schema_compiler/idl_schema.py

Index: tools/json_schema_compiler/idl_schema.py
diff --git a/tools/json_schema_compiler/idl_schema.py b/tools/json_schema_compiler/idl_schema.py
index 8334e7ed8db9b62ca147dd08da568e449f3af3d0..9e22edf7945b924179d5233d088814f3008bcdba 100644
--- a/tools/json_schema_compiler/idl_schema.py
+++ b/tools/json_schema_compiler/idl_schema.py
@@ -88,7 +88,8 @@ def ProcessComment(comment):
 class Callspec(object):
   '''
   Given a Callspec node representing an IDL function declaration, converts into
-  a name/value pair where the value is a list of function parameters.
+  a tuple:
+      (name, list of function parameters, return type)
   '''
   def __init__(self, callspec_node, comment):
     self.node = callspec_node
@@ -96,12 +97,19 @@ class Callspec(object):
 
   def process(self, callbacks):
     parameters = []
+    return_type = None
+    if self.node.GetProperty('TYPEREF') not in ['void', None]:
+      return_type = Typeref(self.node.GetProperty('TYPEREF'),
+                            self.node,
+                            {'name': self.node.GetName()}).process(callbacks)
+      if return_type.get('type', '') == 'object' or '$ref' in return_type:

[not at google - send to devlin, 2013/02/25 18:00:43]

the second argument to .get is unnecessary, it's implicitly None.

[benwells, 2013/02/26 04:19:07]

Done.

+        return_type['optional'] = True;

[not at google - send to devlin, 2013/02/25 18:00:43]

Yes it would be much preferable to fix this by annotating the IDL files directly
for functions that can return null. I imagine this would involve turning

static FileEntry getEntryById(DOMString id);

into

static FileEntry? getEntryById(DOMString id);

I had a quick look at the ppapi source to see if it supports it, but I can't
really tell. Best would just be to do it an see I suppose, but looking at this
file it seem like it'd just be a matter of changing the condition to
return_type.GetProperty('OPTIONAL').

In that case I'd prefer doing it that way.

But if it involves changing the ppapi compiler... what you have is fine, but we
should add a comment.

[benwells, 2013/02/26 04:19:07]

Yep, that would be nicer. The idl parser doesn't seem to like it though. I've
added a comment.

     for node in self.node.children:
       parameter = Param(node).process(callbacks)
       if parameter['name'] in self.comment:
         parameter['description'] = self.comment[parameter['name']]
       parameters.append(parameter)
-    return self.node.GetName(), parameters
+    return (self.node.GetName(), parameters, return_type)
 
 class Param(object):
   '''
@@ -161,8 +169,11 @@ class Member(object):
         properties['description'] = parent_comment
       elif node.cls == 'Callspec':
         is_function = True
-        name, parameters = Callspec(node, parameter_comments).process(callbacks)
+        name, parameters, return_type = (Callspec(node, parameter_comments)
+                                         .process(callbacks))
         properties['parameters'] = parameters
+        if return_type is not None:
+          properties['returns'] = return_type
     properties['name'] = name
     if is_function:
       properties['type'] = 'function'
@@ -226,6 +237,12 @@ class Typeref(object):
     elif self.typeref == 'ArrayBuffer':
       properties['type'] = 'binary'
       properties['isInstanceOf'] = 'ArrayBuffer'
+    elif self.typeref == 'FileEntry':
+      properties['type'] = 'object'
+      properties['isInstanceOf'] = 'FileEntry'
+      if 'additionalProperties' not in properties:
+        properties['additionalProperties'] = OrderedDict()
+      properties['additionalProperties']['type'] = 'any'

[not at google - send to devlin, 2013/02/25 18:00:43]

I think we're going to want a way to actually annotate objects like this as
belonging to the web platform, not an extensions/app schema, but we can do that
later.

+asargent

[benwells, 2013/02/26 04:19:07]

Sounds good.

[not at google - send to devlin, 2013/02/26 05:31:58]

which I suppose would just be adding support for [instanceof=Foo] annotations?

[not at google - send to devlin, 2013/02/26 05:54:20]

I haven't dealt with it much; my understanding was that the instanceOf idiom is
for DOM elements, and maps to instanceof in the JS schema validation. If it
doesn't work then maybe that's unintended, but I don't really know.

     elif self.typeref is None:
       properties['type'] = 'function'
     else: