properties map should be the second element in the vector syntax, not the last.

thanks a lot, yeah silly mistake... but more and more I get more confident I will be able to express any crazy schema + have validating/schema/errors in one shape, thanks for your help

there are no restrictions on reitit for using malli, it should just work. let me test that.

tested and it just works :thinking_face:

if you can do a minimalist repro, I can check what’s wrong with it.

@ikitommi Thanks! Here is the repo (just reused existing mali reitit example and added on top) https://github.com/StankovicMarko/reitit/blob/custom-validation/examples/ring-malli-swagger/src/example/server.clj You can see what i added on top: https://github.com/StankovicMarko/reitit/commit/8f6634d1f7529aebd23e6527ac738857894b0328 Main issue is that when i put values for GET plus, if they are wrong i can't make a request until i put correct value. picture A And for test route i can add whatever i want and i dont want that. I want to prevent it just like with 'native' types picture B

What am i doing wrong?

This was the only i managed to do it

[:map
                            [:date {:description "Date when match happened"
                                    :json-schema/type "string"
                                    :json-schema/format "date"
                                    :json-schema/default "2022-10-10"}
                             [:and
                              [:re #"^\d{4}\-(0[1-9]|1[012])\-(0[1-9]|[12][0-9]|3[01])$"]
                              [:fn {:error/message "Invalid date"}
                               v/valid-date-as-str?]]]]

Am i missing something?

@ikitommi any ideas?

What I have done, is use a deterministic transformation on the generated value in :gen/fmap

that sounds like it might be along the lines of what I'm looking for, I don't suppose you have a code snippet at all?

its not a stupid question, let me see here

(mg/generate
 [:and {:gen/fmap #(str "my_" % "_thing")} string?]
 {:seed 10, :size 10})

something like that work for you?

ah I see, I did try something like that at first but it didn't really work for my situation, let me try and write some pseudocode for what I have

(malli/schema [:fn
                 {:gen/fmap
                  (fn [_]
                    (generate-thing {:seed ???}))}
                 (fn [thing] (validate-thing thing))])

so I'm not sure that would work here, does that make a bit more sense?

can you build up a generator, and supply it in the :gen/gen property?

meaning a clojure.test.check.generators

potentially yes, but this isn't actually in test code, so I'd really only rather use clojure.test.check.generators as a last resort since that would mean putting it in the src imports which, feels a bit funky

if it's not a common thing what I'm doing though I guess I'd have to go down that route

I guess it depends how hard it is to generate your thing

not hard but it does take an int seed

and I'm not sure how to pass that through

since otherwise if I call it without the seed in tests, well, it's random 😅

is it data? it might be worth using malli to describe it

then you will get easier generation

that's a good point, I guess I'm looking for an easy way out as we already have (generate-thing)

how about…

(mg/generate
 [:and {:gen/fmap #(generate-thing %)} int?]
 {:seed 10})

O_o

probably not super good, since the schema is meaningless now

aha I wasn't sure if that would work, as if I use that to validate as well will it try and think the input should also be an int?

yeah

you can use it to generate..

(mg/generate
 [:and {:gen/fmap #(generate-thing %)} 
  [:or int? [:fn ..check for mything..]]
 {:seed 10})

ahh that's interesting

really weird though

yes very weird, 😅 I guess that's how my brain went to trying to get a seed value in at first

yeah not sure it’s super easy

actually.. in:

(malli/schema [:fn
                 {:gen/fmap
                  (fn [_]
                    (generate-thing {:seed ???}))}
                 (fn [thing] (validate-thing thing))])

what would be getting mapped over?

ah nothing, but it does work, I ripped it from the malli test code

this is the relevant example:

(is (= 42 (mg/generate [:re
                            {:gen/fmap (fn [_] 42)
                             :gen/schema :int}
                            #"abc"]))

(https://github.com/metosin/malli/blob/master/test/malli/generator_test.cljc#L171)

but you can generate values from a regex

i guess you can generate values form a :fn schema too

thanks very much for your help btw, I'm about to finish today but you've given me some very interesting ideas

Nice! yeah I am starting on porting a giant codebase from spec and schema (with plenty of custom macros) to malli. Fun times!

I'm in somewhat of a similar situation, it's fun till it isn't has been my experience so far 😅

hope your pulse doesn’t rise too much over it 😛

as an update on this, I ended up writing a custom generator using the clojure test check generators in the end, thanks again for all your help

You have to define your own schema type for that, unfortunately, but it's pretty easy

there is also https://github.com/CrypticButter/snoop

mx/defn should throw when instrumentation is enabled

it does

snoop is cool, but i want to stick to vanilla ways of doing things

my macro just calls instrument at defn creation time

with a specific filter to match only the current function

wanted to keep mx/defn slim. thought of adding the Schema-style :always|never-validate hints too.

how would you implement the never validate mode?

I have a macro emitting the function with an extra {::validate! id} metadata and emitting an instrumentation filter after doing the core/defn call. so that’s a way to do the always-validate behavior

not sure how to guard it, maybe add something into mi/-strument where it filters out on a certain fxn metadata value

relevant bit of :always-validate https://github.com/metabase/metabase/pull/27218/files#diff-e29756aed1e2e2cbce8a26a8e4b2e1cbedfb83cd4d853cf6942455e49516e4d8R44

I assume the 1 arity for coercer decodes or throws, would be nice if we could have 3 arity for on success and on failure continuations, where the on-success takes the decoded value and on-failure takes the explained result

sure, that would work as the arities woudn’t clash.

want to make a PR?

I like the idea. control to the user.

Will do

@ikitommi feedback / ideas before I pursue this path further?

And thoughts regarding adding an epoch transformer?

• cljs: not really sure how to approach that so I just punted and did clj only • will move to experimental for now • namespacing: I thought I did

namespace, so you did, didn’t notice, brilliant! 🙂

I think it would be OK to merge and maybe even release clj-only version first

Though there is some danger in that case that something in the API will be hard to implement on Cljs side

I could explore using date-io as a facade to different JS date libs, so users can use whatever they want: https://github.com/dmtrKovalenko/date-io#projects

Adding clj-only as experimental ns sounds good to me

Personally I’m a bit concerned about the longevity of date-io and would appreciate malli.time to be built upon something that will last over time. However I trust @U061V0GG2 ‘s opinion over mine 🙂

Yeah the issue about Temporal support on date-io shows the the API isn't very well designed

But building our own thing to to support different libs isn't much better either

Besides separating out generators code and providing acceptors for json schema, any other things to keep in mind?

Draft MR is ready, your feedback is appreciated https://github.com/metosin/malli/pull/802

@ikitommi I think it's ready for review, besides adding tests the design is pretty settled. Only thing I'm not sure about is the parsing code which exists in the core ns.

If it did, Given a malli schema, it should return a string with a description of the shape it expects.

Are there problems with doing that? I figure a call to m/ast, then dispatching on :type should be sufficient. wdyt?

you should use m/walk + dispatch on type. Each Schema implements it's own -walk so all the children are walked correctly, see https://github.com/metosin/malli/blob/master/src/malli/json_schema.cljc on how to do this.

but, an interesting idea!

Could you help me understand the benefits of using walk vs ast? Is walk more stable?

All schemas implement the m/-walk. So you can walk over any schema, even those defined in user space. (m/walk schema (m/schema-walker identity)) works always. With AST, you have to know all schemas ahead of time to know how to handle those. In your use case, I guess you need to handle all schemas anyway, you can do the same with AST. Just that it breaks if someone changes the AST for a schema (each schema controls how it looks). There is a disclaimer in Malli README not to use the AST as a persistence model, could change, no no plans to break it and will avoid doing that. We are using AST walking too in projects. But, walk is always safe.

Thanks for explaining that 🙏 I am looking into walk and schema-walker now.

Here’s a self contained example. I used cond since its easier to fit. I’ll actually use a multimethod though

(mc/walk
 [:map [:x :int] [:y :string]]
 (fn [schema _path children _options]
   (println "-------")
   (println "schema:" schema)
   (println "children:" children)
   (cond
     (= (mc/type schema) :int) "INTEGER"
     (= (mc/type schema) :string) "STRING"
     (= (mc/type schema) :map) (str "map of "
                                    (str/join "" (mapv (fn [[k _ v]] (pr-str [k v]))
                                                       children)))
     :else schema)))

I don’t think schema-walker is the way to go, since I dont want to ultimately return a schema

maybe I can attach the stringified explaination to the schema as a property, and look for the prop on children?

I would copy the malli.json-schema and start from that

it's a postwalk, you get the transformed children to your callback, should be straightforward to implement what you want