Strange handling of complement / not about expound HOT 9 CLOSED

BTW I don't think this really matters in practice, though.

clojure.spec is not good at handling bare predicates in terms of error reporting while you can work around it by wrapping them with s/spec:

=> (ex/expound (s/spec (complement str/blank?)) "")
-- Spec failed --------------------

  ""

should satisfy

  (complement clojure.string/blank?)



-------------------------
Detected 1 error
nil
=>

It also works to just s/def a name for the predicate:

=> (s/def ::non-blank (complement str/blank?))
:user/non-blank
=> (ex/expound ::non-blank "")
-- Spec failed --------------------

  ""

should satisfy

  (complement clojure.string/blank?)

-- Relevant specs -------

:user/non-blank:
  (clojure.core/complement clojure.string/blank?)

-------------------------
Detected 1 error
nil
=> 

from expound.

@vemv Thanks for reporting this! I had noticed this as well last night, and you're right, this is really confusing! I'll work on a fix.

from expound.

@vemv Unfortunately, it turns out that we can only do a little better in this case. The slash is really confusing, but the best I can do at this point is to print out "" since spec doesn't capture the original code. Here's what it looks like in my branch (compared to the default spec output for reference):

user=> (require '[clojure.spec.alpha :as s]                                                                                                                               
  #_=>          '[clojure.string :as string]
  #_=>          '[expound.alpha :as expound])
nil
user=>
user=> (s/explain (fn [x] (not (string/blank? x))) "")
val: "" fails predicate: :clojure.spec.alpha/unknown
nil
user=> (expound/expound (fn [x] (not (string/blank? x))) "")
-- Spec failed --------------------

  ""

should satisfy

  <anonymous function>



-------------------------
Detected 1 error
nil

from expound.

I can likely do better once https://dev.clojure.org/jira/browse/CLJ-2068 is fixed!

from expound.

Ah, nevermind:

Functions - demunge the function name and use the qualified function name symbol as the form. Add a special check for anonymous functions and revert to ::unknown for those (not much we can do with an eval'ed anonymous function).

from expound.

OK, I've just released a new version of "0.3.4-SNAPSHOT" which contains these changes (you'll need to delete your old copy in ~/.m2/repository/expound/expound and reinstall)

from expound.

Hey there!

Thank you, much better now 🍻

Perhaps if expound/expound was a macro, it could take the quoted predicate form as a "fallback print value" in case the predicate was unprintable?

I realise turning defns into macros is bit of a dubious move (and a breaking change). Maybe expound* could be offered?

from expound.

@vemv Excellent, I'm glad it works.

You're correct, if expound was a macro, I could do more. However, this might quickly get complicated. For instance, if the call to expound was (expound/expound (s/coll-of (complement string/blank?)) [""]), expound would need to locate the predicate within the spec and match it with the failing predicate.

I suspect it could be done, but since wrapping the predicate in s/spec (or defining a new named spec) seems like a pretty good workaround, I'm going to leave this for now. I may revisit it in the future if this becomes a common pain point.

Thanks for testing the fix!

from expound.

I've just released 0.3.4, which includes this bug fix.

from expound.