@nathanmarz here's my use case, see if you think Specter makes this easier. I'm building a library for modular behaviour trees. i've decided to represent these as nested vectors a la Hiccup markup so, for example [:selector [:loop {:count 99} [:sequence [:randomly {:p 0.5} [:foo]] [:randomly {:p 0.25} [:bar]] [:baz]]]]

what i want to do is (1) label each vector with an autogenerated id, (2) check that required options are specified (e.g. :loop must always have a :count option) (3) check required number of children is specified (e.g. :loop always has exactly 1 child)

i have code doing this based on postwalk (well i've not done 3 yet but it's a trivial modification of 2)

at the moment I am hoisting the options map (and the id) into the metadata for the vector

i'm undecided whether a better approach would be to add an options map to all vectors and put the id in there instead

@sandbags yes, this is exactly the kind of stuff specter excels at

i'll show you an example

(one thing that the walk approach does not seem to make particularly easy is giving diagnostic information)

(def data [:selector [:loop {:count 99} [:sequence [:randomly {:p 0.5} [:foo]] [:randomly {:p 0.25} [:bar]] [:baz]]]])

(def ALL-VECTORS
  (recursive-path [] p
    (if-path vector?
      (continue-then-stay ALL p))))

(let [data2 (transform (subselect ALL-VECTORS META :id) (fn [s] (-> s count range)) data)]
  (select [ALL-VECTORS (collect-one META :id)] data2))
;; =>
[[0 [:foo]]
 [1 [:randomly {:p 0.5} [:foo]]]
 [2 [:bar]]
 [3 [:randomly {:p 0.25} [:bar]]]
 [4 [:baz]]
 [5
  [:sequence
   [:randomly {:p 0.5} [:foo]]
   [:randomly {:p 0.25} [:bar]]
   [:baz]]]
 [6
  [:loop
   {:count 99}
   [:sequence
    [:randomly {:p 0.5} [:foo]]
    [:randomly {:p 0.25} [:bar]]
    [:baz]]]]
 [7
  [:selector
   [:loop
    {:count 99}
    [:sequence
     [:randomly {:p 0.5} [:foo]]
     [:randomly {:p 0.25} [:bar]]
     [:baz]]]]]]

that's adding the auto-generated id to the metadata of each vector, the select call is just so you can see what it did

interesting, thank you

i definitely need to go back and read the docs again

here's how you could insert ids at index 1 of each vector:

(transform (subselect ALL-VECTORS (srange 1 1)) (fn [s] (->> s count range (map vector))) data)
;;=> [:selector 7 [:loop 6 {:count 99} [:sequence 5 [:randomly 1 {:p 0.5} [:foo 0]] [:randomly 3 {:p 0.25} [:bar 2]] [:baz 4]]]]

and as one last example, here's how you would manipulate a particular node type in the tree:

(defn node-type [kw] (path (selected? FIRST (pred= kw))))

(transform [ALL-VECTORS (node-type :randomly) (nthpath 1) :p] #(* 2 %) data)
;; => [:selector [:loop {:count 99} [:sequence [:randomly {:p 1.0} [:foo]] [:randomly {:p 0.5} [:bar]] [:baz]]]]

that's very cute

it's really a shame Specter isn't better known. I know about it and a number of times i've messed around for hours with walk/zippers etc. trying to do something that would have been easier with a bit of Specter

I consider it essential for immutable programming

I've encountered a surprising amount of resistance to it though

i guess one downside (but probably just inherent in this kind of complex structure bending) is that those expressions take some real picking apart

it's always going to make for a bit of a learning hill

I wouldn't consider conciseness a downside, but the opposite

definitely a learning curve

what you're seeing here is the result after me learning this stuff for years

that's what i mean

it's a different way of thinking

but it brings true simplicity and an incredible amount of power over your data

i don't doubt it

what do Rich & co. make of it?

no idea, I'm curious about that myself

they haven't said anything about it to my knowledge

Stuart Holloway, for example, I should have thought would find this interesting.

I suggested contributing specter to clojure awhile back, but the response from core team was "no" with minimal explanation https://groups.google.com/forum/#!searchin/clojure/contribute$20specter%7Csort:relevance/clojure/qN1UPMVQmaM/uegIimKjCwAJ

also saw this recently which makes me think rich would have serious problems with specter: https://gist.github.com/reborg/dc8b0c96c397a56668905e2767fd697f

since he's philosophically against the same function having different performance characteristics on different datatypes, he won't like how I designed the navigators

nthpath for instance being O(n) on lists but faster on vectors (O(log n) I think)

hrmmm

Ah well, perhaps for the best anyway