are compiler-options ignored for compile?

depends?

:optimizations only applies for release

I see. So release means compile and optimize, or is it just a separate step?

all shadow-cljs builds are basically 2-in-1 if you are used to other build tools

compile is one time dev compile, watch is dev compile on repeat, release is a release compile + optimization

got it…I’m going to have to apply my asciidoc skills to your documentation 😉

your documentations skills definitely outshine mine

whenever I start writing docs I start drifting to the code and just write some other code instead

yeah, it’s tough…as you write them you discover bugs and missing things…and have to fix them right then

exactly

I’m definitely liking the i18n ideas…I just discovered that trf is broken on extraction…compiler is moving the literal string to a var

from a plug-ins perspective, wouldn’t it be better just to give a clean API for code to store/retrieve data during compile, and then a hook to run at end of compile?

I love that macros give you the ability to extend the compiler…but it had never occurred to me that lifecycle would also be so useful.

same thing with data_readers, now that I think of it

compiler should also be able to help with css extractions and things like that

could certainly create a proper API instead of directly messing with the cljs.env/*compiler* atom

come to think of it…lifecycle is sufficient

start and end, possibly events in th emiddle

what do you mean by lifecycle? build events?

you have access to the JVM and can create your own state

yeah…:compile-started :compile-finished

right

then an i18n package could just namespace their own atom, init it, and dump at the end

shadow-cljs has these stages currently https://github.com/thheller/shadow-cljs/wiki/Custom-builds#compilation-stages

can't really use an outside atom

it really needs to be in the analyzer data

why not? The JVM is a shared resource

otherwise it won't be restored properly due do caching

shadow-cljs caches compilation pretty aggressively

that breaks whenever you touch things the compiler is not aware of

oh…so you couldn’t just clear it

sure you can clear it but I haven't done a lein clean in years

nice…that’s good to hear 🙂

so yeah, that kind of tooling would need a way to ensure that all files get processed

thats my I'm dumping the strings into the analyzer data per ns

instead of one global collection

can't reliable cache the global thing

come to think of it hot reloading may actually mess up the string collection and collect repeatedly

so on startup the analyzer goes through all files, or is that timestamp shortcut too?

I mean, with caching, how would it ever work to re-run a compile to gather strings?

(other than to clean first)

only the result of the compilation is cached

its a two step process, first all entries are followed and collection

for that only the ns is parsed nothing else

dep graph

then things are sorted in order

and compile

when a given namespace is to be compiled it is first checked if cache is available for that ns

if yes it is restored into the proper places

if not it is compiled as usual

but then you’d skip the macro eval

yes but the data is in the analyzer cache

which is also restored

cache is .js file + analyzer data

oh…that cache is cached?

ok

basically just [:cljs.analyzer/namespaces the-ns] is written to a file so anything thats in there just works

so, if there was an API with docs about the caching, then it’d be usable as a general store for general plug-ins

yes

ok, you write it and I’ll document it 🙂

I’ve got some things I need to percolate on anyway…so writing some docs for your project would maybe keep me from jumping on something too quickly.

probably should check with David first. I'm just abusing the analyzer data here, not sure if that is officially ok 😉

oh..right, that’s in code you don’t control

but given the namespaced keys and general clojure philosopy on open data structures that should be ok

JAM in some keys baby!

the compiler also isn't really concerned with any of it so it really should be ok

I'll write the cljs.i18n things tomorrow

its a good example use case. will ping David to get his opinion with that

I’ll start a book for you tonight if you’d like. Pull together your wiki stuff and start augmenting. asciidoc ok with you?

sure

never used it but your docs look great so I'm in 😉

cool. It’s very similar to markdown

just more powerful and pretty

I added :use-document-host false then what I got was one of the stranger error logs I've seen 🙂 https://pastebin.com/MwwCmCsH short research led me to inline scripts in html causing this, it's maybe a longshot here https://developer.chrome.com/extensions/tut_migration_to_manifest_v2#inline_scripts (I'm also perfectly fine not haveing hot-reload, so no stress about this)

weird, why devtools attempting to load ALL the files

oh right ..

when you generate the single output file the required things aren't set

https://github.com/thheller/shadow-cljs/blob/master/src/main/shadow/build/output.clj#L377-L378

basically that needs to be added somewhere in the file

the karma stuff doesn't do that since its not concerned with hot reload

but yeah the security permissions are weird

maybe you need to allow in the manifest?

it tries to load from localhost which you are probably not allowed to connect to?

I haven't received not-permitted error yet, but I think they include message that x feature need to be permitted. But not ruleing that one out still.

1 sec...

main.js:89433 DEVTOOLS: load JS goog/debug/error.js
main.js:89487 Refused to execute inline script because it violates the following Content Security Policy directive: "script-src 'self' blob: filesystem: chrome-extension-resource:". Either the 'unsafe-inline' keyword, a hash ('sha256-M37Ss2WONZF7RdLxeAaEIHlfH/yox80jtDrUm4wIlr8='), or a nonce ('nonce-...') is required to enable inline execution.

it is trying to load the goog/debug/error.js since it doesn't know that its already loaded

but thats what hot reload would be attempting to do

so that would fail too

ok now it works!

haha, facepalm, I was moveing node_modules folder around and shadow-cljs wasnt installed as dev-dependency. I installed it and now it's connected without warnings.

Or the fact I reloaded the plugin, could be either or.

weird, that shouldn't affect the plugin but who knows

DEVTOOLS: connected!
main.js:89433 DEVTOOLS: disconnected!

[2018-01-11 01:47:54 - WARNING] stale websocket client, please reload client :vast

it may also be complaining due to this trick

https://github.com/thheller/shadow-cljs/blob/master/src/main/shadow/cljs/devtools/client/browser.cljs#L51-L57

that basically tries to execute inline script which is blocked

ok, just one step backward, this is the right way to start the hot-reload in the cmd shadow-cljs cljs-repl :vast ?

so might have to use js/eval instead

where :vast is my id

shadow-cljs watch vast

cljs-repl vast only if you want the actual repl

(while watch is running)

ok, I got all the errors back now again

yeah it is most likely the script-eval thing

it seems to be ok with loading the code via xhr

its just blocking the eval

yes, and the way to permit it is with sha256 hash, I guess that hash changes a lot in development

yeah, maybe eval works

you can copy this file into your project

https://github.com/thheller/shadow-cljs/blob/master/src/main/shadow/cljs/devtools/client/browser.cljs#L51-L57

and let the script-eval just use (js/goog.global.eval code) instead of creating a script element

maybe that just fixes it

ok, I give it a try!

maybe js/eval direct instead of going through goog.global try and see what works 😉

@thheller I’m looking at dev-http. I don’t see a way to plug into the middleware…am I missing anything?

you can only specify a custom handler (via :http-handler)

in short this fixes it, but I haven't yet seen the hot reload in action, could be that I need to look closer at where shadow.cljs.devtools.client.browser is being injected since I'm not useing the same name. Only quirk was that I had to add this into my manifest "content_security_policy": "script-src 'self' 'unsafe-eval'; object-src 'self'"

ah…that’s probably sufficient

I see. I didn’t look above the middleware

@tony.kay you are are providing a server anyways don't you? might be best to just use that instead of the dev http server shadow-cljs provides

working on the book…it’s a thing that is sometimes useful

so scanning through what is supported

ah 🙂

@hlolli I only suggested copying the file into your project to replace the one in shadow-cljs

specifically to only test if js/eval would work

can't test this easily myself since I have no basic setup for a chrome extension

if you have something available publicly I can take a look

ok, I'll zip it and send this to you. Alternatively I could replace this one file, but I'd need to know how to compile shadow-cljs

no you just the file into your classpath

src/shadow/cljs/devtools/client/browser.cljs if you :source-paths is ["src"]

that will then override the default

no need to compile shadow-cljs itself or anything

the classpath ordering will ensure that your version is loaded over the version in the jar

ah didn't know that works 🙂

might need to restart the compile process

but otherwise it just works

no error or warning in the console, I make changes, it compiles and nothing happens. You mentioned it shouldn't try to load all the files, the console looks like this:

yeah that shouldn't happen

basically it should NOT load anything before REPL init successful

it only does because the goog.dependencies_ stuff is not properly set

(because of manually generating the file)

hmm ok, so I could add it?

yes

here

(defn flush-crx-to-file
  [{:keys [polyfill-js unoptimizable build-options build-sources] :as state}
   {:keys [output-to] :as config}]
  (let [prepend
        (str unoptimizable
             (output/closure-defines-and-base state)
             "var shadow$provide = {};\n"
             "goog.global[\"$CLJS\"] = goog.global;\n")
        out
        (->> build-sources
             (map #(data/get-source-by-id state %))
             (remove #(= "goog/base.js" (:resource-name %)))
             (map #(data/get-output! state %))
             (map :js)
             (str/join "\n"))]
    (spit output-to (str prepend out)))
  state)

ok!

yeah add a map step that adds the two lines per file

hmm you could try just adding :devtools {:async-require true} and just use the default :browser output

ie. not a single file

the async-require will load things via fetch instead of the usual closure debug loader

fetch should just work

that would leave me again with the background-script thingy?

I really need to sleep. I can take a look tomorrow

way too tired. gn8.

no, 😄

haha yes good night, sleep well!

is there a recommended “path of least awkwardness” for deploying maven artifacts from a clojure project which is otherwise fully ‘managed’ by shadow-cljs?

sadly that would be using lein (or boot)

I hope someone writes a generic deploy utility library that isn’t coupled to a build tool (ie. also works with tools.deps)

hm. I guess I could have the project.clj read :source-paths and :dependencies from shadow-cljs.edn, https://stackoverflow.com/a/7741790/3421050

avoid duplication of those entries

yeah that works

similar to this https://clojureverse.org/t/combining-tools-deps-with-leiningen/658/4?u=thheller

boom! https://gist.github.com/hlolli/265b9183566a4c5829d7ee355e3d0998 @thheller ❤️

BUJAKA!
main.js:89714 DEVTOOLS: connected!
main.js:89714 DEVTOOLS: REPL init successful
main.js:89714 DEVTOOLS: load JS vast/core.cljs
core.cljs:31 BUJAKA!
main.js:89714 DEVTOOLS: load JS vast/core.cljs
core.cljs:31 BUJAKA!

> I hope someone writes a generic deploy utility library that isn’t coupled to a build tool While I agree that this would be a nice thing, deploying is a somewhat complex task and in the end you’ll end up with something like lein or boot. Now you have three things 🙂

yeah probably

but if boot had a declarative configuration for that I could just call clj -m boot.deploy 😉

declarative as in reading from a file?

yeah, :boot.deploy/config {...} in deps.edn

or separate file

and are you attached to running clj instead of boot? 😉

neither

shadow-cljs clj-run boot.deploy with [boot/deploy ...] in my deps

generic clojure. not coupled to a build tool

lein run -m boot.deploy … everything that is able to launch a JVM should work

I like boot but I don’t like that its all code and not really very declarative

It’s an interesting idea to not have a binary like lein boot and just invoke namespaces

All Clojure build tools were conceived at a time when there was no tool to construct a classpath for you

> I like boot but I don’t like that its all code and not really very declarative It doesn’t have declarative interfaces because those generally end up falling short — that said bolting declarative interfaces on top of boot is something that is very trivial and should probably be embraced more

It would be interesting to see what boot would look like if there was no boot binary

its definitely worth having a dedicated binary still

can do way more optimizations for startup time and such

but that doesn’t necessarily conflict with the thing working as a library as well

https://code.thheller.com/blog/shadow-cljs/2017/11/18/the-many-ways-to-use-shadow-cljs.html

my take on the subject I wrote recently

its definitely more work though so it might not actually be easy to do

too bad clojure makes it so hard to run isolated instances, too much static stuff

java -cp (clj -Spath) boot.App repl running boot without boot 😄

> too bad clojure makes it so hard to run isolated instances, too much static stuff not sure I understand what you mean @thheller

well the reason boot pods exist basically

can’t “fork” a clojure runtime