Improve inline source mapping and error reporting

[arichiardi]

This patch follows some guidelines from Planck and aims at improving source map
consumption in lumo.

The first step was to enable source maps in case of a "load form" (require, ...)
and to refactor make-eval-opts so that it can be reused in repl.cljs. After that the :source-maps key needs to be passed to the cljs.stacktrace machinery for mapping. Additionally, we patch every instance of <embedded> in stacktrace entries so that we can read from the bundled main.js.map file the mapping.

This patch changes this:

1 is not ISeqable
	 Object.cljs.core.seq (NO_SOURCE_FILE <embedded>:506:410)
	 Object.cljs.core.first (NO_SOURCE_FILE <embedded>:507:213)
	 cljs.core.ffirst (NO_SOURCE_FILE <embedded>:567:464)
	 (evalmachine.<anonymous>:1:18)
	 ContextifyScript.Script.runInThisContext (vm.cljs:50:33)
	 Object.runInThisContext (vm.cljs:139:38)
	 (Object.ft)
	 (Object.lumo.repl.caching_node_eval)
	 (NO_SOURCE_FILE <embedded>:5837:273)
	 E (NO_SOURCE_FILE <embedded>:5838:269)

To this:

1 is not ISeqable
        cljs.core/seq (cljs/core.cljs:1223:20)
	cljs.core/first (cljs/core.cljs:1232:16)
	cljs.core/ffirst (cljs/core.cljs:1744:11)
	ContextifyScript/Script/runInThisContext (vm.cljs:50:33)
	vm/runInThisContext (vm.cljs:139:38)

A couple of glitches in ClojureScript where identified and fixed (CLJS-2492 and CLJS-2493) so probably some parts will evolve more in the future but this is mapping fine now! 🎉 👍 🥇

Note that the loading of main.js.map is onerous the first time that it is done and the patching should be considered an experimental feature.

Also big thanks to @mfikes for doing the custom parsing and demunging for planck.

[arichiardi]

CLJS-2493 got applied.

CLJS-2492 did not, but David suggested a different approach for reading/writing source maps I can try

[arichiardi]

I updated the code in order to include planck-repl/planck@27a8e07 and planck-repl/planck@15a89b0 as well.

I tried to compile against the latest master of ClojureScript but I get a weird error with the cache-edn->transit step of the build.

[arichiardi]

There is a new patch for CLJS-2492, waiting for some feedback.

[arichiardi]

CLJS-2492 has been merged! I will rebase this branch on master but I feel #353 should go in first so that it is easier to build against newer ClojureScript versions. What do you think @anmonteiro?

[arichiardi]

@anmonteiro this basically does part of the work you have in the older https://github.com/anmonteiro/lumo/commits/smaps.

It does not have core files and source maps loading but it works pretty well at the REPL because every required namespace already populates the :source-maps compiler key now that the two ClojureScript patches are in.

How about merging and adding core file maps on top? In a separate PR I mean?

[arichiardi]

Rebased.

[arichiardi]

Back into WIP mode in order to add the final hack touch 😄

[arichiardi]

This should be ready to review, I added Mike as well for visibility 👍 but probably things are slightly different in planck regarding source maps.

[mfikes]

LGTM, but hard to tell without running the code.

[arichiardi]

Yep this is actually slowing down things, especially the guessing part. It is there mainly for showing what is possible with some magic (aka - stack trace string parsing).

I was able to get Clojure-like stack traces 😄