@jumblerg sounds interesting

yeah, in theory it should produce a significant improvement.

in practice, we're talking about the browser here, so...

lol, pretty much

hey, so at some point my live reload from reload boot task broke

i don’t really remember when

what are the minimum requirements for that?

i used to see the little cljs logo popup when i made changes, even that has gone now 😞

Reload broke for me when I started using web workers. It appears to inject code into the page that doesn’t work in a web worker context.

Do you see errors in the console?

@candera yeah, reload not compatible with workers

when i was experimenting with web workers i used a separate cljs.edn

the problem is the reload wcode depends on window

which doesn't eexist in the worker context

(the global object is a Worker, not a Window)

i think i have my working setup in git somewhere, i'll check

i had reload going for the non-worker code

I am using a separate cljs.edn, because I have to use simple optimizations in my worker for some reason I can’t remember.

@candera https://github.com/alandipert/lbbasic/tree/b7a7ea1064d6d62b17ba918ea3833d635cc2b8e6

https://github.com/alandipert/lbbasic/blob/b7a7ea1064d6d62b17ba918ea3833d635cc2b8e6/build.boot#L35-L36

Hmm. Apparently no I don’t - works now. Wonder if I changed it chasing the reload thing. Awesome! My program just got faster. 🙂

@alandipert Oh cool!

oh right, using simple optimizations in worker.. also because :none is dependent on window

which kinda stinks because, no source maps in web worker code 😞

giving the experience a very 2012 feel

I just changed it to advanced and it works, though. But yes, I believe that was the reason I switched.

Huh. Reload not working. Weirder, I’m getting an error message about a page other than the one I have loaded.

@candera This was created today: https://github.com/adzerk-oss/boot-reload/pull/110

May solve the reload on webworkers problem.

oh neat! tiny change too

Cool stuff, @mynomoto !

But the problem with reload right now does not seem related to web workers. I’ve punted it once again, as hitting refresh is not a big deal for me.

Latest problem that people had with reload is related to boot 2.7.0 removal of :target-path as explained at https://github.com/adzerk-oss/boot-reload/issues/106

I did a bunch of small things to have reload working and I really like it now. Before I just did what micha says and did a full window reload but I find that flow disturbing now

@mynomoto what kinds of things did you do to make it work?

(ns github-client.core
  (:require
    [devtools.core]
    [github-client.page.core :as page]
    [github-client.reducer :as reducer]
    [github-client.handler.core :as handler]
    [github-client.route :as route]
    [github-client.state :as state]
    [hoplon.core :as h :refer [defelem case-tpl cond-tpl for-tpl if-tpl when-tpl]]
    [hoplon.jquery]
    [javelin.core :as j :refer [cell] :refer-macros [cell= defc defc=]]))

(defonce
  _one-time-side-effects
  (do
    (enable-console-print!)
    (devtools.core/install! [:custom-formatters :hints :async])
    (state/restore-and-watch-db)
    (route/init! state/db)
    (state/start-sync-title)
    (h/do-watch state/title
      (fn [old new]
        (set! (.-title js/document) new)))
    (reducer/start! handler/global state/db reducer/queue)
    (reducer/dispatch reducer/queue [:init])
    ))

(defn init! []
  (reducer/stop! reducer/queue)
  (reducer/start! handler/global state/db reducer/queue)
  (js/jQuery #(.replaceWith (js/jQuery "#app") (page/show {:db state/db :queue reducer/queue}))))

(init!)

It's mostly setup and pointing boot-reload to init!

onetom was my inspiration, some of his project had a similar setup.

what about cells and stuff?

Well every once and while I do a full page reload and my machine has 16GB of ram so not a big problem so far.

i see

Also there is only one cell that has a datascript db created on a defonce, others are formula ones that will create garbage that will not be collected.

But they are scoped on the page/show and there is no risk of stale references I think.

I'm trying a global db for a change, I'm happy with it so far. I think I will be able to push a initial version later today.

Not that I wasn't happy with local cells before but I'm trying to setup a way to debug stuff where users could send me the db, I could load it and see what happened.

@mynomoto: that's very interesting, how do you feel about single atom state?

@flyboarder Well trade-offs as always. If your app don't have views that share data I don't think there is a reason to do it and each view would be happy with local state. But if you can have stale things on a view because it was updated on another and you don't want to make requests each time you change views the global state helps.

Having a central event handler also helps to have one place to debug order of events and if they happened at all. I'm pretty happy with that.

if you have say formula cells that are global or are scoped to some part of your program but accessed in multiple components etc

then i think you no longer have single atom type state

the thing i don't like about single atom state is that you can't have anonymous things in there

How is it no longer a single atom state? Formula cells are derived state no? So the only canonical state there is is the single cell no?

I do have anonymous formula cells in this setup, I don't follow the part about not having anonymous things.

the formula cells don't need to be pure functions

or they can depend on local cells too

etc

they can update even when the atom does not

it's just the idea of having essentially a message bus in the atom that i don't like

Oh, I see. I don't have any other non formula cell in the app.

you don't have components that maintain their own state with input cells?

@micha not in this experiment. Every thing is on the global state.

interesting

I'm still not sure about performance on low end devices but so far it works in my machine which doesn't mean much.

yeah i don't think performance is affected either way

at least intuitively i would think the two approaches would be comparable

since the actual work being done is about the same

like your formula cells are pruning updates still

and your transactor function is like a javelin dosync

i'm interested to see how it affects the composability of your components

my thinking is that local state, anonymous cells, etc facilitate composition by encapsulating state at different levels of abstraction

maybe that's wrong though

with the big atom you have only one level of abstraction

but i suppose you could make some device that uses gensyms to simulate lexical scope

Yeah, conventions are really important with the big atom.

it makes libraries of components somewhat brittle i would think

or at least overly verbose to use

I'm currently using namespaced keywords for identifying local state.

right but when you haev components that interact with each other

they will need to know about each other's keys

this is like programming to concrete types vs interfaces

Well the other way they would need to share cells right? Something is needed anyway.

@micha im trying to get my boot-nodejs server task to restart whenever the server file changes but not have it interfere with the rest of the pipeline

I'm currently trying to figure out when to restart the server, how do I track the one file?

you're using the fileset-diff stuff right?

@flyboarder ^^

I was modeling it after how the hoplon task tracks files