Compiler improvement

[hhugo]

Do no merge, Review only.

• example:
  • rm warning
• compiler
  • parsing (javacript)
    • javascript parser from facebook/pfff ( successfully parsed jquery)
    • more coming ..
  • refactoring
  • cleanup
    • Primitive.mark_used
  • short variable names
    • delay var assignment
    • compute js variables liveness
    • use ocamlgraph to color the graph of js variables (Js_var.ml) (disabled by default)
  • fix
    • remove ouput_file in case of compilation failure
    • profile 3 (-opt 3) : missing deadcode elimination btw inline and flow
  • static evaluation of constant
  • improve parallel renaming by tracking deps.
  • sharing
    • javascript string
    • alias primitives
  • static evalution
  • detect caml_format_int "%d"

[hhugo]

@vouillon, I've pushed a javascript parser (extracted from facebook/pfff).
I had to complete the js ast (with_statement & debugger are still missing).
Can you check printers/precedence in Js_ouput ?

I would like to use this parser during linking.

• compute Requires and Provides (and functional purity ?)
• check javascript files
• better deadcode_elim/inlining
• minify runtime

any suggestion ?

note: I fixed 2 missing semicolon in the runtime. (pushed on master). I don't know if the parser should handle those cases.

[vouillon]

Semicolons are optional at line end, but it's considered good practice anyway.

You can go ahead with using a parser during linking. We should merge this branch first, though. I'm not sure we want to use it to compute Requires and Provides. In particular, how can you tell whether a function is part of the browser APIs or whether it corresponds to a requirement? On the other hand, the three other points make sense.

Also, you don't want to automatically infer functional purity: functions can be implemented in an imperative ways (loops, cached values, ...) while still appearing functional externally.

[vouillon]

The precedences in Js_output seem correct.

[vouillon]

This is an impressive amount of work. You need to fix the -pretty option to generate readable variable name again, and I would disable the variable naming optimization (in Js_var) for now. Otherwise, I'm eager to see this merged.

Thanks a lot!

[Drup]

By curiosity and interest, could we have a comprehensive Changelog of this ? The first post is a bit short on explanations from an outside point of view but there seems to be a lot of interesting new things.

[hhugo]

• variable renaming disabled by default
• CHANGES updated

You may proceed to the merge ;)

[hhugo]

CHANGES may need more entry