[jvm] implement typed functions

[Simn]

Instead of using MethodHandle we now create sub-classes of haxe.jvm.Function, which is similar to what genjava does. This is much faster and more portable, but comes at the expense of increased code size.

There are still some things to optimize in that regard, but I'll do that a bit later.

[Simn]

Some numbers!

Before:

              anon:  59,460,935 (100.00%)
     class-to-anon: 115,746,653 (194.65%)
           dynamic:     819,458 (  1.37%)
     field closure:  62,293,594 (104.76%)
             final: 119,627,458 (201.18%)
          instance: 119,461,624 (200.90%)
         interface: 118,431,625 (199.17%)
     local closure:  61,555,244 (103.52%)
    local function:  61,760,265 (103.86%)
          override: 118,713,814 (199.65%)
            static: 117,879,233 (198.24%)
    static closure:  62,556,824 (105.20%)

After:

  Suite: call with 2 args
              anon: 119,041,326 (100.00%)
     class-to-anon: 119,529,594 (100.41%)
           dynamic:  66,212,033 ( 55.62%)
     field closure: 118,752,228 ( 99.75%)
             final: 118,634,051 ( 99.65%)
          instance: 119,508,294 (100.39%)
         interface: 119,584,280 (100.45%)
     local closure: 119,223,602 (100.15%)
    local function: 119,425,084 (100.32%)
          override: 119,589,525 (100.46%)
            static: 119,544,342 (100.42%)
    static closure: 119,468,944 (100.35%)

nadako-convert before: -- TOTAL  20.43400001525879
 nadako-convert after: -- TOTAL  12.132999897003174

formatter-benchmark before: 22.532000064849854
 formatter-benchmark after: 16.901999950408936

[Gama11]

That's awesome! :)

Some genjava vs genjvm numbers would be interesting too.

[Simn]

genjava:

  Suite: call with 2 args
              anon:  23,959,943 (100.00%)
     class-to-anon:  22,331,613 ( 93.20%)
           dynamic:  22,300,974 ( 93.07%)
     field closure:  14,666,964 ( 61.21%)
             final: 116,229,575 (485.09%)
          instance: 115,997,209 (484.12%)
         interface: 116,688,866 (487.01%)
     local closure:  79,517,529 (331.87%)
    local function:  93,364,918 (389.67%)
          override: 116,328,615 (485.51%)
            static: 116,712,443 (487.11%)
    static closure:     477,268 (  1.99%)

Formatter benchmarks are roughly the same. I get a bit of variance there so it's hard to tell, but benchs.haxe.org will tell us once we merge this.

[kevinresol]

Would be interesting to have a blog post explaining all these