ax3

This is an AS3 to Haxe converter that tries to be very smart and precise about rewriting code. To achieve that, it actually resembles the typical compiler a lot, so here's how it works:

• parse as3 modules into the ParseTree structure, containing all the syntatic information
• load classes and signatures from the SWC libraries so we have the external type information
• process ParseTree and build the TypedTree, resolving all imports and type references and assigning type to every piece of code
• run a sequence of "filters" which analyze and re-write the TypedTree structures to adapt the code for Haxe
• generate haxe modules from the processed TypedTree

DISCLAIMER

This tool was developed and used by InnoGames to migrate our ActionScript 3 codebases. Feel free to ask questions, fork and contribute fixes. However, we are NOT planning to maintain and provide official support for this project.

Usage:

java -jar converter.jar config.json

where config.json is something like:

{
  "src": "<as3 sources>",
  "hxout": "<hx output>",
  "swc": [
    "playerglobal32_0.swc",
    "<other swc libraries>"
  ]
}

Known limitations

• The parser doesn't currently support ASI (automatic semicolon insertion). The only case where a semicolon can be omitted is the last expression of a block.
• Only a small, most commonly used subset of E4X is supported. It's recommended to rewrite the unsupported things in AS3 sources to adapt it for conversion.

Building

The converter is written in Haxe 4 and is only one library, format, for reading the SWC files. To make it easy there's a Lix scope configured, so assuming you have nodejs installed, you can do:

• npm i lix
• npx lix download
• npx haxe build.hxml

It uses the Java target (actually, even the new JVM bytecode target), because JVM has great GC and handles unoptimized functional code of this converter very well. Originally I used JS target, but node.js worked slower and eventually died on so many allocations. :)

TODO

Most of the TODOs are actually in the code, so look there too, but still:

• don't parse *= as a single token when parsing signatures (fix a:*=b parsing without spaces)
• add a "final-step" filter to remove redundant TEHaxeRetypes too
• rewrite arr[arr.length] = value to arr.push(value)
• generate "type patch" files for loaded SWCs, replacing Object with ASObject and * with ASAny
• review and cleanup ASCompat - rework some things as static extensions (e.g. Vector/Array compat methods)
• add some more empty ctors to work around HaxeFoundation/haxe#8531
• add configuration options for some things (like omitting type hints and private keywords)
• fix imports
• add imports for fully-qualified names that can come from @haxe-type
• remove duplicate imports (can happen when merging in out-of-package imports)
• maybe add inline for arithmetic ops in static var inits where all operands are also static inline
• remove public from @:inject/@:postConstruct/@:preDestroy as these should not really be part of public API