Upgrade to ClojureScript 1.10 w/o Google Closure Compiler

[arichiardi]

For now this is what has been done:

• Embed our own build-affecting-options.
• Apply all upstream changes not involving the Google Closure Compiler
• Bundle cljs.core.specs.alpha
• Use cljs.core/eval in lumo.core

[arichiardi]

Unfortunately the commit google/closure-compiler@c668ed7 break us big time because it seems it does not emit goog.provide anymore for compiled outputs.

Bigger time than the JVM implemenation because we don't have access to classes and therefore we cannot hack our own solution.

I don't still grasp all the implications but is seems that

lumo/src/cljs/bundled/lumo/closure.cljs

Line 1454 in 781b5d2

(.split #"(?=goog\.provide\(\")"))]

is not splitting properly into processed modules.

[arichiardi]

Juho pushed another nice fix there: clojure/clojurescript@b1de2f8

Keeping track here just for the sake of it, at the moment these methods are not exposed to the JS compiler.

[arichiardi]

After a night's sleep I think the possible approaches to this are three:

• deviate from core and get dependency info + process some other way
• don't bump the old JS compiler with the current implementation - meh
• have the Closure team expose some dependency info + methods

[arichiardi]

I am trying things at the REPL and what stopped working is the following:

(def compiledCode "var ...")
(closure/parse-js-ns (.split compiledCode #"(?=goog\.provide\(\")"))

which happens here:

https://github.com/anmonteiro/lumo/blob/cljs-1.10/src/cljs/bundled/lumo/closure.cljs#L1454

[arichiardi]

Something I discovered today on this branch is this:

cljs.user=> (doc println)
-------------------------cljs.core/println([& objs])  Same as print followed by (newline)nil

[arichiardi]

New GCC version is out finally solving our long standing source map bug!

I will integrate it here. Wondering if after merging this we should make an alpha release. I am currently using the two outstanding patches merged together into one on node with no issues.

[arichiardi]

Rebased on master in order to have the transit fixes.

[arichiardi]

This PR is stuck because of the compiler issue, still haven't found a way to pass post GCC module processing around. Vanilla Cljs queries the Google Closure Compiler, we can't (see above).

[arichiardi]

Got some debugging of what happens in get-closure-sources with the can-promise library (choosen because tiny):

[[node-module-deps]]
*** module type :es6
*** module type :es6
*** module type :es6
*** module type :es6
js-modules (module$home$arichiardi$git$lumo_repros$node_modules$can_promise$index module$home$arichiardi$git$lumo_repros$node_modules$window_or_global$lib$index module$home$arichiardi$git$lumo_repros$node_modules$can_promise$package_json module$home$arichiardi$git$lumo_repros$node_modules$window_or_global$package_json)
---
var module$home$arichiardi$git$lumo_repros$node_modules$can_promise$index={};var G$$module$home$arichiardi$git$lumo_repros$node_modules$can_promise$index=module$home$arichiardi$git$lumo_repros$node_modules$window_or_global$lib$index.default;
module$home$arichiardi$git$lumo_repros$node_modules$can_promise$index.default=function(){return typeof module$home$arichiardi$git$lumo_repros$node_modules$window_or_global$lib$index.default.Promise==="function"&&typeof module$home$arichiardi$git$lumo_repros$node_modules$window_or_global$lib$index.default.Promise.prototype.then==="function"};var module$home$arichiardi$git$lumo_repros$node_modules$window_or_global$lib$index={};module$home$arichiardi$git$lumo_repros$node_modules$window_or_global$lib$index.default=typeof self==="object"&&self.self===self&&self||typeof global==="object"&&global.global===global&&global||this;
---
parse-js-ns {:requires [], :provides []}
Cannot read property 'replace' of null

	evalmachine.<anonymous>/clojure.string.replace (evalmachine.<anonymous>:2104:72)
	evalmachine.<anonymous>/ (evalmachine.<anonymous>:2840:39)
	evalmachine.<anonymous>/ (evalmachine.<anonymous>:1072:205)
	evalmachine.<anonymous>/cljs.core.LazySeq.fn (evalmachine.<anonymous>:1072:210)
	evalmachine.<anonymous>/cljs.core.LazySeq.sval (evalmachine.<anonymous>:788:151)
	cljs.core/-seq [cljs.core/ISeqable] (evalmachine.<anonymous>:794:255)
	evalmachine.<anonymous>/Object.cljs.core.seq (evalmachine.<anonymous>:502:172)
	cljs.core/-reduce [cljs.core/IReduce] (evalmachine.<anonymous>:793:250)

no more goog.provide even with :whitespace

[arichiardi]

After more investigating, it seems like the options we need have not been exposed to the JS version of the compiler.

Unless someone does it they won't add them there because the GWT has been contributed and seems not really updated.

In particular one thing that is evident from the above is that we need --module-root, see google/closure-compiler-js#96.

This would not solve the problem though of identifying the modules in the GCC-ed file.

It seems like a combination of --module https://github.com/google/closure-compiler/blob/master/src/com/google/javascript/jscomp/CommandLineRunner.java#L223 and --module_wrapper would achieve what we need.

[arichiardi]

Uncommented test-node-modules-cljs-2246 and test-deps-api-cljs-2255 which pass as well with flying colors 🌈

[anmonteiro]

missing moving the generated files for cljs.core.specs.alpha to the target dir

[arichiardi]

Ok I thought I added those, will add them again

[arichiardi]

Adressed.

[anmonteiro]

does adding the code below help you debug what's going on in tests?

[arichiardi]

Yep this is a better reporter basically, using this at work too, can take it away and add it in another PR as well if you want