Using clojure.spec in development

use spec to explicitly check that data structure have the keys/values/structures a function expects

what is the best approach to do this

• validate incoming parameters of functions by explicitly validating the parameters inside the function, e.g. with this utility function:

(defn validate! [spec value error-msg]
  (when-not (s/valid? spec value)
    (s/explain spec value)
    (throw (ex-info error-msg
                    (s/explain-data spec value)))))

Another option seems to be using s/fdef and then stest/instrument to apply upon function call.

What's the preferred approach for data validation during function calls?

seancorfield 20:11
instrument is intended to be dev/test only. For production it's appropriate to use s/valid?

So I should put explicit s/valid? calls inside most of my functions?
Functions at the boundary, not "most".

Also, perhaps, taking more time and care to design the "data API" so you aren't "changing a key name".

Use case
So considering I'm not exactly creating production code, just some code to train a ML model, and still changing functions all the time, it's an ok approach to use instrument? Once the code is stable and using it as an "api", I should get rid of those and validate at the boundaries?

seancorfield
Yeah, absolutely use instrument while you're developing -- that can be very helpful.
I recently had to work on some legacy Clojure code that I'd never worked with before and it was heavily data-centric but I didn't have a sense of the different shapes of data through the pipeline I found it very helpful to write a lot of specs, instrument a lot of functions, and explore it in the REPL. Took me about four or five days to build up a full model in Spec so I could safely begin changing the code 🙂

It can be particularly difficult for beginners with Clojure, given its focus on small, composable pieces -- and composable ideas -- so there is so much "assembly required" when putting together your dev env and your workflow.

Start with the issue's examples of clojure.spec, s/valid?, s/fdef, and stest/instrument, then review the surrounding discussion about development, production boundaries, and REPL exploration. Done means providing a clear, project-appropriate recommendation for when to use each approach, including the boundary-validation guidance.