Skip to content

Latest commit

 

History

History
75 lines (55 loc) · 3.35 KB

README.md

File metadata and controls

75 lines (55 loc) · 3.35 KB
Raw

EchoJS - an ahead of time compiler and runtime for EcmaScript

Building EchoJS

Things only build reliably on OSX. I have easy access to other platforms, I just haven't had the time/motivation to do it given OSX is what I have in front of me every hour of every day, and ios development is my biggest goal. Patches welcome!

You'll need a couple of external dependencies to get things running:

  1. node.js
  2. llvm 3.4
  3. coffeescript

The following commands should get you from 0 to echo-js built:

$ brew tap homebrew/versions  # so we can get the specific version of llvm below
$ brew install node
$ brew install llvm34 --all-targets
$ npm install -g coffee-script
$ npm install -g node-gyp
$ export LLVM_SUFFIX=3.4      # see below
$ export MIN_OSX_VERSION=10.8 # only if you're running 10.8, see below
$ cd echo-js
$ git submodule init
$ git submodule update
$ make

The environment variable LLVM_SUFFIX is used since homebrew installs the llvm34 executables as, e.g. llvm-config-3.4 instead of llvm-config. If you're using llvm 3.4 built from source, you can leave the variable unset.

As for MIN_OSX_VERSION: homebrew's formula for llvm34 doesn't specify a -mmacosx-version-min= flag, so it builds to whatever you have on your machine. Node.js's gyp support in node-gyp, however, does put a -mmacosx-version-min=10.5 flag. A mismatch here causes the node-llvm binding to allocate llvm types using incorrect size calculations, and causes all manner of memory corruption. If you're either running 10.5 or 10.9, you can leave the variable unset. Otherwise, set it to the version of OSX you're running. Hopefully some discussion with the homebrew folks will get this fixed upstream.

both of these variable assignments can be placed in echo-js/build/config-local.mk. I'm running mavericks locally, so my config-local.mk is as follows:

LLVM_SUFFIX=-3.4

But... Why?

  1. I was a PL geek in college, which is pretty much a lifetime ago.

  2. I'd never written a compiler myself, nor played with LLVM, both of which I now can quite confidently say are a blast.

  3. I want to play around with what is essentially profile guided optimization, but with runtime type information. So you get a partially specialized (at least as much as the static compilation can give you) implementation, which then records type information at runtime. You feed this back into the compiler and get a more heavily specialized version.

  4. Right now there's no way to run a JIT on IOS devices. So at the moment the only JS competition for Echo in the use cases I'm envisioning are the JS engines in interpreter mode. When I did my initial testing, spidermonkey was faster than JavaScriptCore, so I've been using SM as the performance goal. It should be possible to beat the interpreter pretty easily, and the PGO/type inference gains should get us up near (but likely not reaching in the general case) the JITs.

Big thanks

Echo wouldn't be as far along as it is now (and certainly wouldn't be as fun to work on) if not for the following:

  1. Esprima: ariya/esprima
  2. Escodegen: Constellation/escodegen
  3. LLVM: http://llvm.org/git/llvm.git
  4. Narcissus: mozilla/narcissus