I defined an fdef spec for a macro and I’m testing it with (is (thrown? Exception (macroexpand ‘(my-macro bad-args))))

when I run it in the repl lein with-profile +test repl the test passes

but when I run it with lein test the arguments are not checked and the test fails

Has anybody hit a similar problem?

is your macro code actually being loaded? macroexpand doesn't throw an error if my-macro doesn't exist and isn't defined as a macro

it was getting loaded

the problem turned out to be related: it was loaded but not recognized by macroexpand because I was not using a syntax-quote

@hiredman thank you for your help

Sometimes s/conform and s/unform are not inlined. For instance:

(s/def ::my-spec (s/cat :op (s/? (s/cat :a integer?))))
(->> (s/conform ::my-spec '(1))
     (s/unform ::my-spec))

There is an extra nesting

Live demo here: http://bit.ly/2efyHkI

Is it a bug in clojure.spec @alexmiller ?

is there a way to get a spec that (s/conform ::my-spec v) will always conform to? I want to fspec a function saying that it takes a spec-destructured value as an argument, and returns another spec-destructured value

did you try s/any??

@sander

I should rephrase, I want a specific spec that only accepts values in the shape that (s/conform ::my-spec v) returns

Interesting question...

@viebel yes, generally anytime conform unform returns a different result, that's a bug. This particular one is already logged.

jira?

@sander no, other than saying what you just said as a custom predicate - call s/valid? on the result with another spec

ok, should be easy enough, thanks!

@viebel http://dev.clojure.org/jira/browse/CLJ-2003

thx @alexmiller

Rich and I talked about this one on Friday in particular - it's tricky

Why is it tricky?

Conceptually tricky about what an optional regex should return in different contexts, how it nests, etc

In my specific case, I was able to workaround this bug using s/conformer. But I still don’t understand fully how s/conformer works. Could you explain in simple words how does s/conformer work?

The docstring is a bit obscure...

It acts as a predicate but applies an arbitrary transformation in the conformed value

What is the simplest use case for conformer?

Applying an arbitrary transformation to the conformed value

It is primarily a tool for building more complex spec variants

It's generally recommended to use sparingly as embedding it in a spec means that you are overriding how spec conforms for all future consumers of your spec

with s/and?

Yes

Like if you used s/or but didn't want a tagged result

When you do that you throw away info about how something was conformed

One use might not need it but other possible future uses may

So baking it into your public spec takes away that option

Thx

I’ll play with that...

@alexmiller another use case would be when one wants to match regex with vectors

The naive code would be:

(s/def ::vec-of-int-and-str (s/cat :0 integer?
                                   :1 string?))

(->> (s/conform ::vec-of-int-and-str [1 "a"])
     (s/unform ::vec-of-int-and-str))

There will be a different solution for that

But yes

but it returns: (1 "a”) instead of [1 “a”]

So the fix is:

(s/def ::vec-of-int-and-str (s/and
                              vector?
                              (s/conformer identity vec)
                              (s/cat :0 integer?
                                     :1 string?)))

Is there a better solution for that?

Not yet but it's coming

I’m curious… ☺️

noob question, i’m sure, re: conformer - i reached for conformer when i was operating on data that looked like “123”, but i wanted (s/conform ::my-spec) to parse the integer such that it ended up being 123. what should i use instead of s/conformer in situations like that, for coercion of this sort?

i felt like i read about something that would solve this problem idiomatically in the rationale or guide, but couldn’t find it when i went back to look for it

@jrheard: yes I have a lot of cases where I have strings as input but I want them conformed to integers (for API input specs). But @alexmiller has told me a couple of times that is "not idiomatic" and to avoid conformers like that :(

hm, i don’t understand - the strings have to be turned into integers at some point, right?

is it more idiomatic to have a secondary, non-spec transformation layer that happens after s/conform?

i’m sure @alexmiller has good reasons for saying that, but i don’t understand what spec users are actually supposed to do here

The downside is that once you enshrine a conversion into your spec, you have made a conformance decision for all future consumers that has removed information. If you understand this ramification and are willing to live with the consequences, then that's up to you.

There are many ways to apply conversions - you should be aware of when you are conflating conformance and conversion though

i think i follow those statements, but a “in situations like that, try doing this instead” example would be supremely helpful 🙂

There's not one answer

You can apply post transformations

You can apply a second spec

You can apply a different spec in one case vs@another

Etc

And in some cases conformers might be the right answer