Usage of $ref causes error under certain circumstances

[RST-J]

While finishing #188 I found out that there is an issue with $ref under certain circumstances.
If you checkout the branch any_ref_issue from my fork and run the tests you'll see

  1) Failure:
UriRelatedTest#test_schema_ref_from_file_with_spaces [/home/jonas/repos/json-schema/test/test_uri_related.rb:59]:
{"names"=>[{"first"=>"john"}]} should be valid for schema:
{"$schema"=>"http://json-schema.org/draft-04/schema#", "type"=>"object", "required"=>["names"], "properties"=>{"names"=>{"type"=>"array", "items"=>{"anyOf"=>[{"$ref"=>"test/schemas/ref john with spaces schema.json"}]}}}}.
--- expected
+++ actual
@@ -1 +1,4 @@
-[]
+["The property '#/names/0' of type Hash did not match one or more of the required schemas. The schema specific errors were:
+
+- anyOf #0:
+    - The property '#/names/0' was not a valid schema"]

I tracked it down to this spot where the schema is registered with full file URI but the determined schema key is just a path (which should be absolutized as in normalized URI).

(byebug) JSON::Validator.schemas.keys
["fd615350-b519-5597-a476-eb93ab9046b6#", "file:///home/jonas/repos/json-schema/test/schemas/ref%20john%20with%20spaces%20schema.json#"]
# but
(byebug) schema_key
"test/schemas/ref john with spaces schema.json#"

I feel like normalized_uri should be used here too. Has anyone a quick idea of how to do that?

P.S.: And why has the registered schema's key a # at the end?

[pd]

The registered schemas all get an empty fragment because we use uri.fragment = '' when we create them:

uri = Addressable::URI.parse('http://example.com/hi')
uri.to_s #=> "http://example.com/hi"
uri.fragment = ''
uri.to_s #=> "http://example.com/hi#"

If we assigned nil instead, we wouldn't have the empty fragment. I've considered changing it before but didn't really understand the consequences, so left it alone. =)

[RST-J]

@pd so, you suggest that we never register schemas with a hash rather than keep that and looking them up with adding one if it is missing?

[iainbeeston]

I'd have thought that, because $ref's for a schema always end in '#' it would be simpler to store them with that. Also, surely that way is optimised for the common case? (lookup is faster, at the expense of an extra operation during storage)

[RST-J]

TBH, I was not sure about which way to go (and speculated you'll interfere if I pick the wrong choice), with or without # and wondered about which would be faster (which is the only criterion here I think).

Is it really required that $ref's end with #?

To increase performance, it would be necessary to reliably determine when a normalization is required and when not, so that we can skip calling addressable if an URI already is normalized. But I have no idea how to do that.

[iainbeeston]

@hoxworth might be a good person to ask about this. Or I could poke to common test suite guys, who are usually good at pointing out the correct way to interpret the spec

[RST-J]

Sure, if they can give us any useful hints we appreciate that.

[iainbeeston]

I decided to post it on the json-schema google group first (rather than raising an issue on the common test suite just to ask a question)

https://groups.google.com/forum/#!topic/json-schema/9KdnjylLp8k

[BGarretson]

I'm having this problem; can someone confirm whether or not this bug was supposed to have been fixed in 2.5.0 or if the fix is yet to be released?

#214