Skip to content

Latest commit

 

History

History
101 lines (83 loc) · 3.43 KB

README.md

File metadata and controls

101 lines (83 loc) · 3.43 KB
Raw

wasm

WebAssembly for Scala

Provide some AST and a way to convert them to wasm binary. There is also a printer for binary wasm.

There is two way to use wasm, either as a cli or a library.

install

It's using sbt for building, simply ensure that you have it.

To ensure that it's working, you may want to run sbt test prior to using wasm.

cli

It can be used directly from the sbt prompt, there is no system-wide binary installation.

library

To add to the local ivy storage

sbt publishLocal

and add libraryDependencies += "default" %% "scalawasm" % "0.1" to your build.sbt

use

cli

To compile wat file to wasm file

sbt run compile watfile1 watfile2 ...

To print the content of a wasm file

sbt run print wasmfile1 wasmfile2 ...

library

The main target is to be able to have a library to generate wasm, usable in scala, from a compiler perspective.

The compilation pipeline and data type used at each stage is

pipeline         text.Lexer -> text.Parser -> binary.ToBinaryAst -> binary.ToBinary
datatype    String  ->  ast.Token   ->   ast.Tree     ->     ast.Binary   ->   Stream[Byte]

The most suitable coding entrypoint is wasm.ast.Tree, it provide a convenient, yet nearly binary, tree.

A good code starter is

for {
  ast <- Module(
    name = Some("My Module"),
    funcs = Seq(
      Function(
        sig = Signature.Function(),
        instrs = Seq(
          Opcode.Nop,
      ))))
  p <- ToBinaryAst(ast)
} yield ToBinary(p)

todo

short

  • ToBinaryAst.Section.Export works only for function, it's trivial to port it to supports others
  • ToBinaryAst.getSections could use some refactoring, to nicely separate which spaces can be used in which section
  • find a good macro system for Scala, to reduce some value duplication in LEB128.Type

medium

  • there is structural support for some wasm extension, such has multi return or multi memory, but it still have to be defined in spec and fully implemented/tested
  • Data section is implemented until Tree but not in the remaining pipeline
  • Custom section is not implemented, there is partial structural support for it
  • remove many warning by using dedicated case object instead of having a generic way to do it

long

  • parser is too simple, it support only "regular" wat, but the spec documentation is not very precise either, that would need some improvement based on the new version of the spec having a better parser would lead to very easy testing, as drop-in of files would works, we could reuse directly the core testing suite of the spec repo (there is still some issues, such as them using some undefined opcode)
    • correctly handle multi lines comments
    • parse the full syntax
    • int/float values are badly handle, it will break on some corner cases, also, the parsing is based on naïve understanding (previously undefined); see the "toInt" TODOs
    • there is some structural check we can push to parser, such as i32 can't load a value bigger that the actual available range
    • order of module opcode is either not well defined or wrong
  • Printer can be much cooler than that and try to regenerate the wat input and having a way to print wasm and also a decompiler
  • support other compilers than the reference one
  • remove quirks used for spec compatibility, but that would require that WebAssembly/spec#625 be fixed first; check for Config.enableSpecCompat