rspec fails on ruby-head
Closed this issue · 5 comments
Currently most of the tests fail on ruby-head. I'm currently not sure why, but I noticed this as my own gem started to fail on ruby-head recently.
What I observed is that it seems the recursive aspect is no longer working. There is nothing in the ruby changelog that I can pinpoint to be causing this, but I may be overlooking something.
If time permits, I'll investigate further myself but I'm not sure if that is going to happen any time soon.
Apparently ruby 2.2.3 has significantly refactored ostruct, in such a way that it breaks the current reverse-open-struct accessor method definitions. At least one factor that plays a role is the defintiion of respond_to_missing? in ostruct, causing the test "unless self.respond_to?(name)" to always return true.
Changing the test into "unless self.methods.include?(name.to_sym)" improves things, but still several tests fail.
The remaining errors, seem to be related to the fact that initialize_copy no longer calls new_ostruct_member any more. I'll prepare a pull request.
Finally traced the issue down to the following commit: ruby/ruby@3bf9b2f. It's not part of the released 2.2.3, but as it is a commit of oct 15, 2015 it is part of 2.2.3dev, which is what Travis currently uses for ruby-head.
The issue with the initialize_copy has a couple of consequences:
- the tests (and the delete_field method) asume that if a key exists in the table, the accessor methods exist.
- calling an accessor on the duplicated ROS, OpenStruct's method_missing will be called, which will generate the accessor methods and return @table[key] instead of calling the newly created method.
- if the first call is _as_a_hash, it will consider "key_as_a_hash" as the key and generate methods "key_as_a_value", "key_as_a_value=" and "key_as_a_value_as_a_value".
We can go different ways here:
- call the new_ostruct_member for each key in the table in intialize_copy.
- patch the ROS to implement lazy creation of attributes, as implemented in the latest OpenStruct and let delete_field call remove_method, but catch and ignore the exception.
The first is the easiest one to implement, but generates overhead for ruby < 2.2.3dev. The thread in pull request 1033 indicates that there can be a significant performance gain by using lazy accessor creation, so I opted for the second solution.
@aetherknight thanks for the merge and release. I noticed my travis tests started passing on ruby-head this weekend ;)