cljs2go

command module
v0.0.0-...-9d268a4 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: May 2, 2016 License: EPL-1.0 Imports: 3 Imported by: 0

README

cljs2go

Way to Go! -- Siobhan Sharpe

[2015-08-01] Was based on ClojureScript 0.0-2411 in 2014, and passes (most of) the tests, though the compiler is JVM based. Should be redone on top of self hosted ClojureScript 1.7 at some point.

ClojureScript to Go. Implemented as an overlay onto ClojureScript, instead of a fork. Go is emitted from cljs.go.compiler, which is a patched version of cljs.compiler. At run time, a thin JS compatibility is provided to avoid having to touch cljs.core, these js and goog packages aren't intended for end-user usage.

Once the compiler starts working, the plan is to provide a second monkey patch in spirit of the first, but this time for the Java dependencies, like java.io.File and their Clojure counterpart, clojure.java.io. This patch is to be able to compile the compiler itself to Go. This won't change the run-time characteristics of the compiled programs, nor will it introduce eval, it's simply done to leverage the Go build pipeline. ClojureScript has a reader in cljs.reader that potentially can replace tools.reader. Macros used via :require-macros needs to be compiled to Go first, and an ad-hoc wrapper for the compiler depending on them will be created to compile the actual source (the basic compiler has the cljs.core macros built in).

While not introducing eval, it does open up for adding it as a third layer, the problem lies on the Go side, as you need a way to build and link code into the running process. This is certainly doable in various ways, but not yet sure what the right way to do so is. One potential way is to use Go's own cgo to both call and expose Go packages as C combined with dlopen, potentially using go-ffi. Another is via net/rpc. A third interesting option is to model the namespaces using go9p. Regardless the approach, it will need a way to keep or migrate the state of vars. There's also Go Execution Modes which hints at a future -buildmode=plugin, potentially in Go 1.5.

Why?

Mainly for fun and to learn Go. But it is also trying to address issues around fast compilation and start-up time (a current concern in the Clojure world, see below). On the other side, see Fernando's motivation for his corvus LLVM Lisp: "The era of dynamic languages is over. There is currently a 'race to the bottom': Languages like Rust and Nimrod are merging the world of low-level programming with high-level compiler features." The reason to start this exploration with Go itself is most pragmatically explained in Rob's motivating talk. And, Go has momentum.

The goal isn't to compete with the JVM or V8 on performance. Another obvious reason would be to leverage the Go ecosystem using ClojureScript. It's also meant to simplify creating new emitters (at least for myself) for ClojureScript. I'm aware of tools.analyzer etc., but they don't help you implement clojure.core and the actual language run-time. That said, building on a fork of tools.analyzer.js instead of cljs.analyzer is a likely future direction. Currently, ClojureScript's analyzer is not modified, and a lot of extra mess happens in the Go compiler. Some of this could / should be moved to analyzer passes.

How?

The easiest way to start understanding how the compiler works is to look at the emitted Go code. core.go is core.cljs compiled to Go. overrides.go are Go specific overrides compiled from overrides.cljs. rt.go is a handwritten Go file providing the implementation needed, mainly AFn and wrappers around Go reflect capabilities and some coercing functions (like Truth_) used by the emitted code. ClojureScript functions are represented by this AFn struct, which bundles up potentially more than one Go function into a ClojureScript one. deftype compiles to Go structs, defprotocol to Go interfaces. Protocol methods are real Go methods, not AFns. number, boolean, array, string and seq are recognized and compiled to float64, bool, []interface{}, string and CljsCoreISeq. Like in JavaScript there are no real integers, all numbers are float64 for simplicity, but this is likely to change.

As can be seen on ISeq above, types and protocols are ns-prefixed to not clash with functions (like Symbol vs. symbol). Public functions in Go must start with an uppercase character. Functions starting with a _ (munged from -) have an X in front of them. The arity is appended, so a full compiled name will look like this: X_invoke_Arity1. ArityVariadic is a special case which regardless of how many fixed parameters take a single varargs parameter which is then unpacked inside the generated body. This is to simplify dispatch and avoid having 20+ different varargs signatures (this might change). Functions with primitives (up to 3 arguments) are compiled into something like Arity1FF (takes one float64 and returns a float64). These functions live beneath the normal protocol IFn dispatch. A normal (without primitives) bridge IFn function is also generated (using MakeFunc) for the matching arity. Like in ClojureScript, there's a special protocol Object that allow creating of methods that look like real Go methods (no _ArityN). These methods, like any host methods or functions, are invoked by the dot notation, like (.toString x). There's currently no way to create a plain Go func, but this will likely become a macro.

When compiling, the unaltered cljs.analyzer from ClojureScript is used to build the AST (see Nicola's AST Quickref). The analyzer depends on the cljs.core macros (core.clj), which (like all ClojureScript macros) are written in Clojure. This file is replaced by cljs.go.core, and heavily uses the js* macro to emit literal Go code. Once the AST has been generated, it's fed into the cljs.go.compiler, which emits the Go source code. This is in turn (usually) fed into goimports (a version of gofmt that also fixes the imports) and then finally go build (or go install).

Usage

Not ready yet.

Ensure you have Go 1.6.x installed and GOPATH setup properly, see How to Write Go Code. Then clone this repo like this:

# Clone and build the Go packages:
$ go get github.com/hraberg/cljs2go
# The repo itself lives under src by convention:
$ cd $GOPATH/src/github.com/hraberg/cljs2go

# go test, for Go tests checked into git, both generated and handwritten ones:
$ go test -v ./...
# To re-generate the tests from ClojureScript (this might dirty the repo):
$ lein test
# (This also generates and runs some tests under target/generated which aren't checked in.)

# To re-generate the Go for ClojureScript itself:
$ go generate

The tests use testify partially forked from da775f0, living under vendor.

While the compiler more or less works, and passes most of ClojureScript's test suite, it's not packaged for actual use, as I first intend to compile it to Go. If you want to play with it, it's easiest to fire up a REPL and look at the cljs.go namespace.

The Hello World sample can be built like this:

;; Generate the Go source from a Clojure REPL:
cljs.go> (compile-file "samples" "samples/hello.cljs")
# go run, < 1s
$ cd samples/hello
$ go run main.go
Hello World

# go build
$ go build -o hello main.go
# Native binary, < 50ms, 7.5M
$ ./hello
Hello World

Road Map

We're currently in phase 1. cljs.core-tests now pass.

Phase 0
  • Generate compiling ClojureScript with mapping for all basic constructs to Go.
  • Minimal JavaScript host libraries assumed by ClojureScript.
Phase 1
  • Getting the relevant parts of ClojureScript's test suite to run.
  • Ability to build an app with more than one namespace and a main fn which depends on the core.
  • Go interop, func macro, coercions functions, ability to write moderately useful programs.
Phase 2
  • Ability transform core.clj, analyzer.clj and compiler.clj to compiling ClojureScript.
  • Compiling one reader, either cljs.reader (which might need extensions) or tools.reader to ClojureScript.
  • Bootstrap the compiler using the above to Go.
  • Ability to build new wrapping compilers for require-macros to work. Assume the macros are valid ClojureScript.
Phase 3
  • Performance. Revisit all the basic assumptions. Generate cleaner code, remove redundant constructs.
  • Move to the tools.analyzer stack.
  • Full bootstrap, porting the hacks that allow the compiler to compile itself.
  • core.async using real Go blocks and channels.
  • ClojureScript to Go REPL
Phase 4
  • Porting to other targets.
See Also
Lisp's in Go:
Dialects of Clojure
Ports of ClojureScript
Lower-level Lisps
Other potential targets

License

cljs2go

Copyright © 2014 Håkan Råberg

Distributed under the Eclipse Public License either version 1.0 or (at your option) any later version.

ClojureScript

Copyright (c) Rich Hickey. All rights reserved. The use and
distribution terms for this software are covered by the Eclipse
Public License 1.0 (http://opensource.org/licenses/eclipse-1.0.php)
which can be found in the file epl-v10.html at the root of this
distribution. By using this software in any fashion, you are
agreeing to be bound by the terms of this license. You must
not remove this notice, or any other, from this software.

Documentation

The Go Gopher

There is no documentation for this package.

Directories

Path Synopsis
cljs
core
Go overrides.
Go overrides.
reader
Go overrides.
Go overrides.
clojure
data
Non-core data functions.
Non-core data functions.
set
Set operations such as union/intersection.
Set operations such as union/intersection.
string
Go overrides.
Go overrides.
walk
This file defines a generic tree walker for Clojure data structures.
This file defines a generic tree walker for Clojure data structures.
zip
Functional hierarchical zipper, with navigation, editing, and enumeration.
Functional hierarchical zipper, with navigation, editing, and enumeration.
js
samples

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL