start of ES6 modules: move to rollup

[alicebob]

This is the minimal version of #178

The build process is still to concat the JS files (as ./dist/js/pzpr.concat.js), but this file is now a ES6 module. This concated file is then transformed for the browser via rollup. The actual code of the module didn't change. The tests do use the modern import ... from ... syntax to load data.

[robx]

Mostly a nitpicky review with whatever came to mind while first reading over it. Feel free to ignore parts that don't make sense. Also I still need to have a bit of a look at rollup.

[robx]

the git-hash part here seems like it should be redundant with the new Makefile target

[robx]

could we get rid of this script entirely and call npx grunt default from the Makefile instead?

[alicebob]

p.html and the examples and such are still copied with Grunt. I thought is was a bit too much to also change that in this PR. But we could.

[robx]

I didn't mean to get rid of Grunt, I meant to get rid of the "build" script in package.json, and to call Grunt directly from the Makefile.

[robx]

is this (still) used anywhere?

[alicebob]

It's referenced in Gruntfile:

  var PRODUCTION = (grunt.cli.tasks.indexOf('release') >= 0);

But the only difference seems to be that if PRODUCTION is true it copies over .map files, which is fine anyway?

[robx]

It's at least fine in the sense that I'm currently deploying those anyway... https://puzz.link/js/list.js.map

[alicebob]

@robx thanks for the comments, this should be it.

[robx]

Having a bit of trouble working with this. Specifically,

$ make build
[...]
npx rollup -c ./rollup.config.js
Unexpected token import
make: *** [rollup] Error 1

Any clue?

Also, we might want to keep Grunt for concatenation, since that's what gives us source maps and thus decent stack traces on test failure. (This works, when run via npx grunt concat:pzpr instead of the bundle Makefile target: e15d00c.)

[robx]

I've pushed two fixes to master that should obviate the need for those isFreeze checks.