Rakudo.js update - we passed a first roast test
After spending a long time chasing a bunch of bugs (a lot of ones that boiled to a few lines of code had really weird consequences that needed to be painstakingly chased down) and implementing some missing parts I finally got Test.pm6 to compile and load.
This allowed rakudo to pass few roast tests (like S03-operators/not.t) for the first time. Once some remaining issues with the Test.pm6 get solved hopefully a whole bunch of others will pass.
Currently the coroutine ops needed for Test (for gather/take type stuff) to load are implemented in a node.js specific manner using the fibres. I previously had an pure js implementation of them using a CPS transform but it carried a large cost in terms of generated code and compile time so it got disabled as it was slowing down development. Some careful thought will be needed later on how to re-nable it with only paying for it for use.
Now that rakudo.js is starting to run actual Perl 6 code my current focus is to make rakudo.js easier for others to play it.
That includes means making the way it builds (at least remotely) sane and cutting down the horrible (10+ second locally) startup cost of loading the setting. A lot of the startup cost seems to be caused by the way lexical scopes are setup in rakudo.js which also needs a cleanup as it's currently to influenced by how nqp does things.