I am getting this exception "Additional data must be non-nil." when using st/check on my function. I have gotten this before and then it mysteriously went away. I am now getting it constantly and am curious what is going on. It is coming from these Spec lines: https://github.com/clojure/spec.alpha/blob/5228bb75fa10b7b13ea088d84f4d031cce74a969/src/main/clojure/clojure/spec/test/alpha.clj#L278-L284.

if you have a reproducible example case, please paste it @kenny

Working on that. It's pretty coupled to this code but I should be able to pull something out.

cool -- thanks!

this has been reported before, but AFAIK there's not a deterministically reproducible case

Hopefully I can get one! It's definitely reproducible for me right now.

the error occurs when you try to create an ex-info with nil data. the case where this happens here is when spec validation fails, but explain-data succeeds (which ideally should never happen). usually, this is a bug in spec. occasionally, it's a bad predicate in user code.

This error is quite confusing. Seems like a bug either way 😉

well the error should never happen

right, it's an invariant violation

haha, the best kind of bugs

I pulled some code out to get a repro case and was running st/check on it. It failed with a regular specification error. I thought that was weird because it hasn't failed yet. Turned on instrumentation and ran again and got the "Additional data must be non-nil." message. May be related to instrument.

this won't help you, but I've made a change in spec-alpha2 to better report this case

Alright, here's a repro: https://gist.github.com/kennyjwilli/f324b94eaadc404cb72fdfe41067b469. Not minimal but I've gotta go make some food. Within 3 or so runs of (st/check formula->fn)`, it will get the exception.

so I misdiagnosed above - even more subtle... s/conform and s/valid? are returning different answers when run on the same input

which is because the spec is an fspec, which generates its own source of random inputs

this is a known issue and generally we've been suggesting people don't use fspec because it has all these unexpected consequences

fspec has proven to be quite valuable though. What is the alternative? Just fn?

ifn?

unfortunately it's not even easy to override in a spec alias or anything

Won't that just generate any old function during gen tests and pass it to my function?

yes, it is significantly harder to use it in check. I'm not saying any of this is good.

this is a known area I'm hoping to spend some rethink time on in spec 2, but we're going to overhaul the function return value stuff first

Oh ok. Any workaround for now to avoid this error?

I don't think I have any simple advice for you, other than to remove your :ret spec :(

:ret on the fspec of the top level fn?

no, the :ret on the fdef

that is an fspec

I was trying to add some gen improvements to clojure.spec.alpha but I cannot load the ns in a REPL. I get

CompilerException java.lang.Exception: #object[clojure.spec.alpha$and_spec_impl$reify__2171 0x4a332076 "[email protected]"] is not a fn, expected predicate fn, compiling:(/home/kenny/Forks/spec.alpha/src/main/clojure/clojure/spec/alpha.clj:78:1) 

hi gents! How I can define spec of vector where I know only first element and don’t know length of vector. For example [:in 1 2 3 4]

it would be great to use something like

(s/tuple #{:in :or} & nat-int?)

(s/cat :in #{:in :or} :nums (s/* nat-int?))

this is exactly the job the regex ops are made for

I see, thank you!