Defer dereferences of metaconstants in function fakes

[philomates]

Fix for #159

Idea behind the fix is to, for function fake results ((foo) => ..some-metaconstant..), defer the dereference of variables pointing to metaconstants. These dereferences were occurring at macro-expansion time due to the use of constantly. By deferring to runtime, we are able to dereference these variables in scopes that have been properly manipulated by =contains=> data fakes.

[philomates]

I noticed that (intern *ns* symbol (Metaconstant. symbol {} nil)) was being run multiple times for the same symbol. Figured it makes sense to filter out duplicates.

[philomates]

the two future-facts are tests elaborating some nice-to-have behavior that currently isn't supported

[rafaeldff]

What would happen in the following:

(fact 
  (:header (gen-doc)) => "gamma"
  (:header (gen-doc)) => "beta"
  (provided
    (gen-doc) =streams=> [..doc1.. ..doc2..]
    ..doc1.. =contains=> {:header "gamma"}
    ..doc2.. =contains=> {:header "beta"})) 

[philomates]

Aah, nice example! It breaks the code, though I had to tweak it to be valid midje DSL code because provided only applies to the fact immediately above it:

(fact 
  (vector (:header (gen-doc)) (:header (gen-doc))) => [1 2] 
  (provided 
    (gen-doc) =streams=> [..doc1.. ..doc2..] 
    ..doc1.. =contains=> {:header 1} 
    ..doc2.. =contains=> {:header 2}))

results in

FAIL at (form-init5605701913344457958.clj:1)
    Expected: [1 2]
      Actual: (nil nil)
       Diffs: in [0] expected 1, was nil
              in [1] expected 2, was nil

I'll add this as a future-fact test for future implementation

[rafaeldff]

Would the following maintain the same behavior as now?

(fact 
  (identical? (gen-doc) (gen-doc)) => true
  (provided
    ..doc.. =contains=> {:header "gamma"}
    (gen-doc) => ..doc..)) 

What about this:

(fact 
  (identical? (gen-doc) (gen-doc)) => true
  (provided
    ..doc.. =contains=> {:header (rand)}
    (gen-doc) => ..doc..)) 

?

[philomates]

These continue to pass. Before they passed because (identical? nil nil) is true, now they pass because calls to gen-doc dereference ..doc.. after it has been bound to {:header (rand)}

[rafaeldff]

Cool. Do you think it would be worth adding this test to the suite?