:asset-path is only used in a few places for JS files, you just specify an absolute path, so :asset-path "/js" or so

any other "assets" you control yourself via HTML or code, so just emit absolute paths there

I guess my question is then "is there a natural way to communicate server URL from shadow-cljs.edn to my frontend code?"

why would it need to? are your assets served by a separate server?

otherwise just use an absolute path, ie. starting with a /

Sorry. I just fixed the issue by doing exactly what you suggested. Thanks for the help!

yes, that would be a bug

not sure what is happening there though

@U05224H0W related, but not necessarily the same issue: I suspect that at least cljc files on the source path that are intended to and also practically only used for self-hosted CLJS (i.e. :target :bootstrap) are run through the Clojure compiler during reload when a watch is active for the normal ClojureScript build (`:target :browser`), and then complain about, for example, no such namespace: js. These files work as intended in the self-hosted environment. Is this intended to work differently? Can I have self-hosted CLJS-only files in my sources that can assume to be run only in a CLJS context?

one build never affects another

the browser build will only compile files it required itself. note that it is extremely easy to make mistakes with .cljc files, thus I rarely recommend using them

bootstrap will also compile more files than other builds do, so it might be deep in some macro files that otherwise are never compiled

my guess is that a macro file somehow fails to load but the error is lost somewhere

maybe because of some ns form that is not valid in clj

and then the macro ns is not defined and leading to the above error

I'm still at a stage where I want to make specter work in an self-hosted environment. For that I'm trying to understand how macros work in the self-host environment provided by shadow-cljs :bootstrap. For that I'd like to step through the code and look at stuff. Not sure how to do that though. I'm on emacs and use cider.

there is nothing special to macro compilation. the ns is just compiled as normal, using :cljs branches in .cljc files

I believe you, but that doesn't yet match the mental model I have built based on my observations so far 🙂 Something must be different though because js/some.name.space.js and js/some.name.space$macros.js are different

I don't know what you mean by that. you should under no circumstance ever refer to any CLJS namespace via js/

if you mean the output files then yes, there are 2. can't write two namespaces into one file

the way you should approach this is that it is basically two entirely separate namespaces, that just happen to have the same name from the code side (so same ns) but different internally

I mean the output files, yes. The two are different, but they are produced from the same file. So something must be different about the compilation process.

Another way to state the problem is that (macrovich/deftime (defmacro ...)) macros don't make it to the $macros output file for the namespace in my situation currently. Also I read that defmacro is ignored – or produces no output – when compiled in CLJS, maybe that's one difference?

More close to my original problem, which is that select is undefined in CLJS in a self-hosted environment, I seems clear now why the upstream version behaves like this - it has (defmacro select ...) in #?(:clj. A PR/fork/branch that was reportedly working better has it in (macrovich/deftime ...) instead. However, select still didn't appear in the ...$macro.js output file. I assumed that that must be correct, but maybe it isn't. Reading https://cljs.github.io/api/cljs.core/defmacro now, that looks helpful.

Hmm, when the cljs js compiler asks for the source of a namespace as a non-macro namespace, shadow-cljs effectively (modulo analyze caching and so on) offers output generated form .cljs or .cljc files, and when asked for a macro namespace, it offers .cljc or .clj, is that accurate? Does that effectively mean that for bootstrapped / self-hosted cljs we can put CLJS code in .clj files?

Also, if I have code that should run at compile-time, and that code is different for cljs jvm (clj compiles macro ns) and cljs js (cljs compiles macro ns) and I want separate files for macro ns from non-macro ns, I end up with a .clj file with #?(:cljs reader conditionals, correct?

can you point me to the actual sources?

The original specter source code for the select macro: https://github.com/redplanetlabs/specter/blob/master/src/clj/com/rpl/specter.cljc#L347 And the PR: https://github.com/jeff303/specter/blob/bootstrap/src/clj/com/rpl/specter.cljc#L359

In the PR the effective change was replacing #?(:clj with (macrovich/deftime.

I have a local repo that shows that defmacros in macrovich/deftime in cljc files don't end up in the compiled js and are thus not available for source code to be compiled by the self-hosted compiler.

Based on quickstart-browser, which I was about to push to github

deftime seems useless to me here, as is the conditional. just remove it entirely, so the code is always loaded

I mean I can't tell if there is any actual java interop in there, but otherwise if its just clojure code just keep it always active

the defmacroalias is also a problem

can't alias vars like that for cljs

Yeah, there's a separate solution for that in the PR that I'll take.

Thanks, I'll try that. I still need to understand it though 🙂 So defmacros` in macrovich/deftime aren't supposed to end up in the JS output?

not a clue. I've never used macrovich, don't know what it means

the problem is that #(:clj ..) hides it from the CLJS macro compilation which will take the :cljs branch. so thats why select is missing, it is never there in the first place

(defmacroalias richnav macros/richnav) don't know why this is there

why not just have the macros directly use macros/richnav. I never get these aliasing things, so not sure why it has to be in this ns

deftime is just 1 line:

(defmacro deftime
  "This block will only be evaluated at the correct time for macro definition, at other times its content
   are removed.
   For Clojure it always behaves like a `do` block.
   For Clojurescript/JVM the block is only visible to Clojure.
   For self-hosted Clojurescript the block is only visible when defining macros in the pseudo-namespace."
  [& body]
  (when #?(:clj (not (:ns &env)) :cljs (re-matches #".*\$macros" (name (ns-name *ns*))))
    `(do ~@body)))

that sounds like it should be what you want

I thought so, but defmacro s defined in a macrovich/deftime don't appear in JS output produced by a shadow-cljs :bootstrap build.

Ah, I may have to put the namespace in :macros in the config.

again, this does not seem like its needed to me

shadow-cljs already ignores defmacro in .cljc files unless compiling macros, so having them there doesn't do anything "extra"

I understand that I don't need deftime here, but I'd like to understand why it doesn't match my mental model. You said before that deftime sounds like it should be what I want. But nothing that's inside deftime ends up in the JS output. Is that expected, given its definition above?

I cannot say without seeing the full actual code

the defmacro should definitely be in the generated $macros.js file, but not in the regular ns .js file

Alright, let me push something to github to show you

@U05224H0W https://github.com/shadow-cljs/quickstart-browser/compare/master...mbertheau:quickstart-browser:master

I just noticed a possible misunderstanding. I didn't mean that the deftime macro itself doesn't end up in the JS, but that what's inside a call to the deftime macro doesn't end up in the $macros.js file.

goog.provide('friday.myns$macros');
console.log("hello from toplevel");
friday.myns$macros.toplevelfunction = (function friday$myns$macros$toplevelfunction(){
return console.log("hello from toplevel function");
});
var ret__5824__auto___14156 = friday.myns$macros.toplevelmacro = (function friday$myns$macros$toplevelmacro(_AMPERSAND_form,_AMPERSAND_env){
console.log("hello from toplevelmacro expansion");

return cljs.core.sequence.cljs$core$IFn$_invoke$arity$1(cljs.core.seq(cljs.core.concat.cljs$core$IFn$_invoke$arity$2((new cljs.core.List(null,new cljs.core.Symbol("js","console.log","js/console.log",-2005248266,null),null,(1),null)),(new cljs.core.List(null,"hello from toplevelmacro result evaluation",null,(1),null)))));
});
(friday.myns$macros.toplevelmacro.cljs$lang$macro = true);

console.log("hello from usetime");

friday.myns$macros.usetimefunction = (function friday$myns$macros$usetimefunction(){
return console.log("hello from usetimefunction");
});

var ret__5824__auto___14160 = friday.myns$macros.usetimemacro = (function friday$myns$macros$usetimemacro(_AMPERSAND_form,_AMPERSAND_env){
console.log("hello from usetimemacro expansion");

return cljs.core.sequence.cljs$core$IFn$_invoke$arity$1(cljs.core.seq(cljs.core.concat.cljs$core$IFn$_invoke$arity$2((new cljs.core.List(null,new cljs.core.Symbol("js","console.log","js/console.log",-2005248266,null),null,(1),null)),(new cljs.core.List(null,"hello from usetimemacro result evaluation",null,(1),null)))));
});
(friday.myns$macros.usetimemacro.cljs$lang$macro = true);


//# sourceMappingURL=friday.myns$macros.js.map

seems to be all there for me?

Not deftimemacro though

oh what is usetime then? confusing names 😛

just saw some time stuff and thought thats it 😛

could be that *ns* isn't actually bound to the $macros ns? never checked

Is that a peculiarity of shadow-cljs bootstrapped cljs? shadow-cljs does compile the cljs to js outside the browser. Is that different in this regard compared to compiling all the namespaces in the browser?

the macro compilation is not done for regular builds, so this is different

I pushed a commit that logs (ns-name *ns*) in all the places and only ever get null or markus.pumpkin, never markus.pumpkin$macros.

*ns* is definitely bound to the correct value. most of your additions were incorrect though. I don't understand the problem with deftime either, but I do not see the point anyways so just do not use it

I guess I can workaround it by not using reify on promesa library and replace it with a deftype, but still is very strange that a method definition of some reify object (that is conditionally initialized on first call) is relocated to other module and called on module load+eval time

known problem

https://clojure.atlassian.net/browse/CLJS-3207

work arround is to set :compiler-options {:cross-chunk-method-motion false} in the build config

thanks! \o/