context is that i am looking for the least-invasive way to get reagent’s macros to work in self-host

since they are part of the public api, it’s not an option to move them to other macro-specific namespaces

should be fine to have them separate. even .clj should be fine as long as it doesn't do anything java specific?

eg. reagent.core/with-let and reagent.ratom/with-let should both remain in the same ‘place’ api-wise

i thought so, i’ve just been running into super weirdness with my latest attempts

eg https://github.com/mhuebert/shadow-bootstrap-example/blob/80707045c9a5739237eaef3b76bd2c57fa8a8072/src/userland/macros.cljc#L4

The comment there is not exactly accurate as I’ve found its nondeterministic

compiling/recompiling I get different results, macros are not always resolved when I expect them to be

so i was wondering if the cljs & cljc passes could be clobbering each other somehow, or the compiler state for the namespaces is being corrupted

reagent’s macros currently do not work on selfhost, for reasons I don’t understand. I also thought that .clj namespaces should be ok as long as they don’t do interop

but we get a ra is not defined error here: https://github.com/reagent-project/reagent/blob/master/src/reagent/core.clj#L10

;; removing this ns causes `m2 is not defined` error

what does this mean? works just fine for me?

also if you comment / uncomment and it reloads a couple of times?

after I posted that I discovered that it does work sometimes

ah thought you get it during compile

ah. no, happens when evaluating in selfhost

recording… it maybe has nothing to do with whether that line is commented

just causing it to recompile, is nondeterministic

there is also this - https://github.com/mhuebert/shadow-bootstrap-example/blob/8ca524a5794ba42f9cbaec71c6121b4db2a4f6c6/src/userland/macros_2.cljc#L2

I was surprised to find that if I would :require-macros a cljc namespace that does not self-require, I can’t use the macros from within it

it never even attempts to load the macro ns when in self hosted

dunno why

ah nevermind it does load everything as expected

it might be that some of the bootstrap index data does not look exactly like the self-host compiler needs it to look

because its generated by shadow-cljs and some of the ns-related metadata looks different

ah I know ...

the CLJ-side dependencies aren't followed when constructing the bootstrap index and compiling macros

at least thats how I remember it. so everything in the macro ns itself is directly loaded but its dependencies are not.

so macro namespaces themselves are fine but macro namespaces with dependencies don't work

hm. I don’t recall running into this before

I’ve written this failing test - https://github.com/clojure/clojurescript/compare/master...mhuebert:cljc-macros-require-test

(defmacro current-ns-str
  "simple macro that returns info only available at compile time"
  []
  ;; no-op is just to test re-use of macros from other namespaces
  `(userland.macros-2/no-op
     ~(name (.-name *ns*))))

this seems to work. using the fully qualified symbol to the alias doesn't need to be resolved.

I don't know why though.

but I suppose it might just end up calling the no-op as a function which in this case is fine

definitely a head scratcher 😛

indeed