s/merge makes me so happy.

being able to just mix in key sets for various stages of the pipeline

beautiful.

the go type system has “struct embedding” which i fell in love with when working in go - lets you accomplish something similar by merging struct fields

i’m not quite sure yet i grok the every vs coll-of, or every-kv vs map-of stuff yet

is the general idea to use coll-of and map-of by default and only switch to every if your specs are too slow and you don’t need conformance

?

also - i’m now running in to a situtation where a spec is not-conforming and it’s just consuming a ton of memory producing the explanation data

@bbloom I haven’t run into that situation, but that’s my guess

yeah, i get the feeling explain needs to become lazy somehow - not necessarily with thunks, but maybe with something like a … that you can ask it to expand further

hmm, yeah - i have a structure with LOTS of structural sharing and a ton of redundant data - not unlike the clojurescript ast

and explain is starting to become non-useful due to the shear volume of output

and seemingly exponential time to return

or print or whatever

not quite able to pin down a reproduction yet

You can set a custom explain printer if that becomes useful

Just setting print-length will affect the default one though

In case that helps

Re your collection question, yes that's a good first approximation

re: print-length - cool! thanks.

You can also set coll-check-limit to further tweak what every and every-kv check

suuuuuppper nit-picky feedback:

if that first-approximation is true, that you should default to coll-of or map-of, then it seems like the more informative doc-strings should be on those, rather than every and every-kv

feel free to totally ignore i even said that 😛

Rich always thought people should default the other way

But in general most people seem to disagree :)

preferring the sampling kind?

Yeah

hm, i was going back and forth on it in my head

seeming like sampling is the right thing if the specs are not intended to be used for contract-style checks

which seems true

but it feels wrong to get non-determinstic behavior by default

I would actually prefer that the doc for both covered all the details, but you know

Rich isn't big on duplication like that

the slippery slope of excessive doc strings - i like the short strings

I have a spec reference page in the works for the docs which will help too

i wish the see also was metadata tho, so the cross referencing was more automatic

and maybe even let you see-also a spec

so like the options that are specified could be generated from common doc strings

If wishes were fishes something something

heh, indeed - the simple strings are generally fine

Later, shutting down for takeoff

aaaaannny way, i guess i’ll lean on the sampling ones & see how that feels

cya, have a good flight

How do you spec a multi-method? Multi-spec seems like it should do it, but it only talks about spec’ing a hash with types. More specifically I have the following - but it never validates the :box parameter.

(s/fdef colliding?
  :args (s/cat :invader map? :box ::box/box)
  :ret boolean?)

(defmulti colliding? :character)

(defmethod colliding? ::small [invader box]

You can't right now

Probably fixable but I haven't looked at it yet

Oh really?

Kinda shocked - I had assumed it was me the whole time.

Might wanna make it clearer in the docs that it’s not supported. All I could find was some confusion on multi-spec.

Thanks for the help tho!

Yeah, multimethod, protocols, primitive typed fns, and inlined fns will not work right now

Inlined and protocols are probably not easily fixable

user=> (binding [s/recursion-limit 1 s/coll-error-limit 1 s/coll-check-limit 1] (clojure.test/run-tests ‘ambiparse.calc-test)) That normally returns in ~1 second, but when instrumented, it seems to be infinitely looping. jstack says: https://gist.github.com/brandonbloom/dfa559f81ad1b6d0980c120d86a2e18e

something seems to not be respecting the limits

or i’m missing a limit

amusingly, i’m working on a parser, soooo i’m running in to exactly the same kinds of infinite loop problems in the program i’m trying to debug 😉

Hi - I'm experimenting with spec and am getting an error that I'm not understanding

RuntimeException Var clojure.test.check.generators/large-integer is not on the classpath  clojure.spec.gen/dynaload (gen.clj:21)

looking at the spec impl, it seems like there isn’t on spec checking level/depth - which might be what i want/need, not sure tho

It's entirely possible some spec impl is not respecting the limit

@derwolfe: you need test.check on your classpath

@alexmiller thanks!