Clojurians Log v2

Q: I’ve switched to an M1 mac/Big Sur and I’m seeing a delay in shadow compiles when using the :watch. Has anyone else seen this?

I use Cursive and “Save All” to initiate. I’m pretty sure Cursive is saving the file because when I try the same thing in a node/browsersync project, the rebuild is instant

I’ve switched to :watch verbose but don’t see anything there.

Any suggestions on how I can diagnose where the delay comes from?

Did the watch verbose include a timestamp?

I just checked: no timestamp

Hey, I have a bit of a weird bug that appeared when I switched from figwheel-sidecar (0.5.20) + cljsbuild (1.1.8) over to shadow-cljs (2.11.18). My application is running on Electron (5.0.2, can't update) and re-frame (0.10.9). The app uses a touch screen and no other peripherals, and the users need to input text, so a homebrew on-screen keyboard is used. Everything worked fine with figwheel, but when I switched to shadow, with no other dependencies being updated or changed, text inputs broke. When the user types into a field with the on-screen keyboard, the caret stays stuck to the start of the input. Naturally I have access to a keyboard when developing, which works just fine for text input: the caret moves as I input characters into the text field. After I've input some text with a keyboard, and switch to using the on-screen keyboard, the caret also continues to behave normally. Only when I input into an untouched field, with the on-screen keyboard, does the caret get stuck. The text field content is managed from re-frame db, and the on-screen keyboard "inputs" append the pressed key to the selected field's state in the db. I've tried to dispatch the same events a physical keyboard would send (`input`, keyup and keydown). I've tried to use .setSelectionRange to manually correct the caret to the right place, but it gets overwritten, so right now the ugly workaround I have is to set the caret position on a short timeout. This leads to the caret visibly jumping as you type, but the characters mostly go where they need to go. I want to stress that this all worked completely normally with figwheel, and only upgrading to shadow broke it. Are there any guesses to what might have broken it? Different optimizations, transpiling to a different version? I might have to switch back to figwheel if I can't solve this, because the current hack to type correctly is very unreliable and fragile.

I can't give the project sources over, but here's the project.clj for the frontend. Shadow config included because of lein-shadow

(defproject xxx "3.3.1"
  :dependencies [[org.clojure/clojure "1.10.1"]
                 [org.clojure/clojurescript "1.10.764"]
                 [org.clojure/spec.alpha "0.2.187"]
                 [medley "1.3.0"]
                 [re-frame "0.10.9"]
                 [garden "1.3.10"]
                 [day8.re-frame/http-fx "v0.2.0"]
                 [com.taoensso/tempura "1.2.1"]
                 [com.andrewmcveigh/cljs-time "0.5.2"]
                 [thheller/shadow-cljs "2.11.18"]]

  :plugins [[lein-garden "0.3.0"]]

  :min-lein-version "2.5.3"

  :source-paths ["src/clj" "src/cljs" "src/cljc"]

  :clean-targets ^{:protect false} ["resources/public/js/compiled" "target"]

  :garden {:builds [{:id           "screen"
                     :source-paths ["src/clj" "src/cljc"]
                     :stylesheet   xxx.css/screen
                     :compiler     {:output-to     "resources/public/css/garden.css"
                                    :pretty-print? true}}]}

  :profiles {:dev  {:dependencies [[binaryage/devtools "1.0.2"]
                                   [day8.re-frame/re-frame-10x "0.3.7-react16"]
                                   [day8.re-frame/tracing "0.6.0"]]
                    :plugins      [[lein-kibit "0.1.8"]
                                   [lein-shadow "0.3.1"]]}
             :prod {:dependencies [[day8.re-frame/tracing-stubs "0.6.0"]]}}

  :npm-deps [[create-react-class "15.6.3"]
             [react "16.8.6"]
             [react-dom "16.8.6"]
             [react-flip-move "^3.0.4"]
             [react-highlight.js "^1.0.7"]]

  :npm-dev-deps [[shadow-cljs "^2.11.18"]]

  :shadow-cljs {:lein   true
                :builds {:app {:target           :browser
                               :modules          {:app {:entries [xxx.core]}}
                               :output-dir       "resources/public/js/compiled"
                               :asset-path       "/js/compiled"
                               :dev              {:closure-defines {"re_frame.trace.trace_enabled_QMARK_"        true
                                                                    "day8.re_frame.tracing.trace_enabled_QMARK_" true}}
                               :devtools         {:preloads   [devtools.preload
                                                               day8.re-frame-10x.preload]
                                                  :watch-dir "resources/public"}
                               :compiler-options {:optimizations :simple}}}})

For one you are not using the correct CLJS version. That should be 1.10.773 for 2.11.18.

what CLJS version did the figwheel variant use?

the cause of this problem will be somewhere in your code but I cannot say where. I vaguely remember some things breaking because of a CLJS update sometime ago but can't remember details

CLJS version was not changed, so it was the same when I used figwheel. I will update CLJS and report back!

Bumping CLJS version to 1.10.773 did not change behaviour. I'm not ruling out a problem in re-frame/reagent/react that manifested by changing build tools, but I'm at loss so I will ask on #re-frame

yeah I don't have a clue what could be causing this

is this with release only or also with development builds?

I've narrowed it down just now to re-frame not firing off the input events until after the virtual keyboard has been closed the first time for each field or physical keyboard input is given. Not sure why this surfaced now, but that's the issue

So I will continue the discussion on re-frame channel if I get stuck again 👍

Hello! Is it possible to require a ns only if goog.DEBUG is true? In Clojure I would use requiring-resolve but that is not available in cljs. (I have a library under my :dev alias dependencies that I use during development but do not want to include in production code.) Thank you!

@holyjak :preloads [that.ns] in your :modules or :devtools

Thank you! I have now added my.ns as a pre-load and in my code I use

(when goog.DEBUG (my.ns/my-fn))

yeah you can't do that

Ok. So what is a solution when I need to call to the dev-only code from my prod code (but only while running in dev)? Perhaps in the preload I could store the fn in some *dyn-var* that lives in code that is save to share between both and call (when *dyn-var* (*dyn-var*)) in my prod code?

I guess I could simply define js/myThing in my dev code and use that from my prod code...

I cannot answer that without more information. I do not know what you are calling this way and what the code actually does. You cannot dynamically require anything so you'll have to find a different solution.

I figured it out. Instead of calling (holyjak.fulcro-troubleshooting/troubleshooting-render-middleware) in my prod code I will call (js/holyjak.fulcro_troubleshooting.troubleshooting_render_middleware) As long as I only do this in dev, it will work.

can't you just create a ns that you add to preload and that adds the middleware in question? or does it have to be in your app ns?

The problem is that it cannot "add the middleware". The prod code has to ask for the middleware. There is, in prod code, something like:

(setup-app {:middleware <some middleware...>})

Has anyone seen this kind of error before? A case if works on my local machine, but stopped working on CircleCI. The build user is root so I am not sure why it would fail a file system operation.

nil
IOException: Operation not permitted
	java.io.UnixFileSystem.canonicalize0 (UnixFileSystem.java:-2)
	java.io.UnixFileSystem.canonicalize (UnixFileSystem.java:178)
	java.io.File.getCanonicalPath (File.java:620)
	java.io.File.getCanonicalFile (File.java:645)
	jdk.internal.reflect.NativeMethodAccessorImpl.invoke0 (NativeMethodAccessorImpl.java:-2)
	jdk.internal.reflect.NativeMethodAccessorImpl.invoke (NativeMethodAccessorImpl.java:62)
	jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke (DelegatingMethodAccessorImpl.java:43)
	java.lang.reflect.Method.invoke (Method.java:566)
	clojure.lang.Reflector.invokeMatchingMethod (Reflector.java:167)
	clojure.lang.Reflector.invokeNoArgInstanceMember (Reflector.java:438)
	shadow.build.node/flush-optimized/fn--18657 (node.clj:224)
	shadow.build.node/flush-optimized (node.clj:212)
	shadow.build.node/flush-optimized (node.clj:209)
	shadow.build.targets.node-library/flush (node_library.clj:194)
	shadow.build.targets.node-library/flush (node_library.clj:189)
	shadow.build.targets.node-library/process (node_library.clj:213)
	shadow.build.targets.node-library/process (node_library.clj:196)

odd. is that permanent or just temporary? maybe some disk stuff going on?

What could this mean? I have done this for ages and never seen this error - "missing instance" - before:

Connecting to remote nREPL server...
Clojure 1.10.1
(shadow/repl :main)
Execution error (ExceptionInfo) at shadow.cljs.devtools.server.runtime/get-instance! (runtime.clj:11).
missing instance

Inside a :builds target, I had :output-dir "./" :output-to "./handler.js This stopped working for some reason. I changed it to :output-dir "output" :output-to "output/handler.js` and it worked! Removing the trailing / does not make a difference. Must be CircleCI or my updated Docker build image...