mirror of
https://github.com/seejohnrun/haste-server.git
synced 2024-11-01 11:31:22 +00:00
2011 lines
87 KiB
JavaScript
2011 lines
87 KiB
JavaScript
/***********************************************************************
|
|
|
|
A JavaScript tokenizer / parser / beautifier / compressor.
|
|
|
|
This version is suitable for Node.js. With minimal changes (the
|
|
exports stuff) it should work on any JS platform.
|
|
|
|
This file implements some AST processors. They work on data built
|
|
by parse-js.
|
|
|
|
Exported functions:
|
|
|
|
- ast_mangle(ast, options) -- mangles the variable/function names
|
|
in the AST. Returns an AST.
|
|
|
|
- ast_squeeze(ast) -- employs various optimizations to make the
|
|
final generated code even smaller. Returns an AST.
|
|
|
|
- gen_code(ast, options) -- generates JS code from the AST. Pass
|
|
true (or an object, see the code for some options) as second
|
|
argument to get "pretty" (indented) code.
|
|
|
|
-------------------------------- (C) ---------------------------------
|
|
|
|
Author: Mihai Bazon
|
|
<mihai.bazon@gmail.com>
|
|
http://mihai.bazon.net/blog
|
|
|
|
Distributed under the BSD license:
|
|
|
|
Copyright 2010 (c) Mihai Bazon <mihai.bazon@gmail.com>
|
|
|
|
Redistribution and use in source and binary forms, with or without
|
|
modification, are permitted provided that the following conditions
|
|
are met:
|
|
|
|
* Redistributions of source code must retain the above
|
|
copyright notice, this list of conditions and the following
|
|
disclaimer.
|
|
|
|
* Redistributions in binary form must reproduce the above
|
|
copyright notice, this list of conditions and the following
|
|
disclaimer in the documentation and/or other materials
|
|
provided with the distribution.
|
|
|
|
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER “AS IS” AND ANY
|
|
EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
|
PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER BE
|
|
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
|
|
OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
|
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
|
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
|
|
TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF
|
|
THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
SUCH DAMAGE.
|
|
|
|
***********************************************************************/
|
|
|
|
var jsp = require("./parse-js"),
|
|
slice = jsp.slice,
|
|
member = jsp.member,
|
|
PRECEDENCE = jsp.PRECEDENCE,
|
|
OPERATORS = jsp.OPERATORS;
|
|
|
|
/* -----[ helper for AST traversal ]----- */
|
|
|
|
function ast_walker() {
|
|
function _vardefs(defs) {
|
|
return [ this[0], MAP(defs, function(def){
|
|
var a = [ def[0] ];
|
|
if (def.length > 1)
|
|
a[1] = walk(def[1]);
|
|
return a;
|
|
}) ];
|
|
};
|
|
function _block(statements) {
|
|
var out = [ this[0] ];
|
|
if (statements != null)
|
|
out.push(MAP(statements, walk));
|
|
return out;
|
|
};
|
|
var walkers = {
|
|
"string": function(str) {
|
|
return [ this[0], str ];
|
|
},
|
|
"num": function(num) {
|
|
return [ this[0], num ];
|
|
},
|
|
"name": function(name) {
|
|
return [ this[0], name ];
|
|
},
|
|
"toplevel": function(statements) {
|
|
return [ this[0], MAP(statements, walk) ];
|
|
},
|
|
"block": _block,
|
|
"splice": _block,
|
|
"var": _vardefs,
|
|
"const": _vardefs,
|
|
"try": function(t, c, f) {
|
|
return [
|
|
this[0],
|
|
MAP(t, walk),
|
|
c != null ? [ c[0], MAP(c[1], walk) ] : null,
|
|
f != null ? MAP(f, walk) : null
|
|
];
|
|
},
|
|
"throw": function(expr) {
|
|
return [ this[0], walk(expr) ];
|
|
},
|
|
"new": function(ctor, args) {
|
|
return [ this[0], walk(ctor), MAP(args, walk) ];
|
|
},
|
|
"switch": function(expr, body) {
|
|
return [ this[0], walk(expr), MAP(body, function(branch){
|
|
return [ branch[0] ? walk(branch[0]) : null,
|
|
MAP(branch[1], walk) ];
|
|
}) ];
|
|
},
|
|
"break": function(label) {
|
|
return [ this[0], label ];
|
|
},
|
|
"continue": function(label) {
|
|
return [ this[0], label ];
|
|
},
|
|
"conditional": function(cond, t, e) {
|
|
return [ this[0], walk(cond), walk(t), walk(e) ];
|
|
},
|
|
"assign": function(op, lvalue, rvalue) {
|
|
return [ this[0], op, walk(lvalue), walk(rvalue) ];
|
|
},
|
|
"dot": function(expr) {
|
|
return [ this[0], walk(expr) ].concat(slice(arguments, 1));
|
|
},
|
|
"call": function(expr, args) {
|
|
return [ this[0], walk(expr), MAP(args, walk) ];
|
|
},
|
|
"function": function(name, args, body) {
|
|
return [ this[0], name, args.slice(), MAP(body, walk) ];
|
|
},
|
|
"debugger": function() {
|
|
return [ this[0] ];
|
|
},
|
|
"defun": function(name, args, body) {
|
|
return [ this[0], name, args.slice(), MAP(body, walk) ];
|
|
},
|
|
"if": function(conditional, t, e) {
|
|
return [ this[0], walk(conditional), walk(t), walk(e) ];
|
|
},
|
|
"for": function(init, cond, step, block) {
|
|
return [ this[0], walk(init), walk(cond), walk(step), walk(block) ];
|
|
},
|
|
"for-in": function(vvar, key, hash, block) {
|
|
return [ this[0], walk(vvar), walk(key), walk(hash), walk(block) ];
|
|
},
|
|
"while": function(cond, block) {
|
|
return [ this[0], walk(cond), walk(block) ];
|
|
},
|
|
"do": function(cond, block) {
|
|
return [ this[0], walk(cond), walk(block) ];
|
|
},
|
|
"return": function(expr) {
|
|
return [ this[0], walk(expr) ];
|
|
},
|
|
"binary": function(op, left, right) {
|
|
return [ this[0], op, walk(left), walk(right) ];
|
|
},
|
|
"unary-prefix": function(op, expr) {
|
|
return [ this[0], op, walk(expr) ];
|
|
},
|
|
"unary-postfix": function(op, expr) {
|
|
return [ this[0], op, walk(expr) ];
|
|
},
|
|
"sub": function(expr, subscript) {
|
|
return [ this[0], walk(expr), walk(subscript) ];
|
|
},
|
|
"object": function(props) {
|
|
return [ this[0], MAP(props, function(p){
|
|
return p.length == 2
|
|
? [ p[0], walk(p[1]) ]
|
|
: [ p[0], walk(p[1]), p[2] ]; // get/set-ter
|
|
}) ];
|
|
},
|
|
"regexp": function(rx, mods) {
|
|
return [ this[0], rx, mods ];
|
|
},
|
|
"array": function(elements) {
|
|
return [ this[0], MAP(elements, walk) ];
|
|
},
|
|
"stat": function(stat) {
|
|
return [ this[0], walk(stat) ];
|
|
},
|
|
"seq": function() {
|
|
return [ this[0] ].concat(MAP(slice(arguments), walk));
|
|
},
|
|
"label": function(name, block) {
|
|
return [ this[0], name, walk(block) ];
|
|
},
|
|
"with": function(expr, block) {
|
|
return [ this[0], walk(expr), walk(block) ];
|
|
},
|
|
"atom": function(name) {
|
|
return [ this[0], name ];
|
|
}
|
|
};
|
|
|
|
var user = {};
|
|
var stack = [];
|
|
function walk(ast) {
|
|
if (ast == null)
|
|
return null;
|
|
try {
|
|
stack.push(ast);
|
|
var type = ast[0];
|
|
var gen = user[type];
|
|
if (gen) {
|
|
var ret = gen.apply(ast, ast.slice(1));
|
|
if (ret != null)
|
|
return ret;
|
|
}
|
|
gen = walkers[type];
|
|
return gen.apply(ast, ast.slice(1));
|
|
} finally {
|
|
stack.pop();
|
|
}
|
|
};
|
|
|
|
function dive(ast) {
|
|
if (ast == null)
|
|
return null;
|
|
try {
|
|
stack.push(ast);
|
|
return walkers[ast[0]].apply(ast, ast.slice(1));
|
|
} finally {
|
|
stack.pop();
|
|
}
|
|
};
|
|
|
|
function with_walkers(walkers, cont){
|
|
var save = {}, i;
|
|
for (i in walkers) if (HOP(walkers, i)) {
|
|
save[i] = user[i];
|
|
user[i] = walkers[i];
|
|
}
|
|
var ret = cont();
|
|
for (i in save) if (HOP(save, i)) {
|
|
if (!save[i]) delete user[i];
|
|
else user[i] = save[i];
|
|
}
|
|
return ret;
|
|
};
|
|
|
|
return {
|
|
walk: walk,
|
|
dive: dive,
|
|
with_walkers: with_walkers,
|
|
parent: function() {
|
|
return stack[stack.length - 2]; // last one is current node
|
|
},
|
|
stack: function() {
|
|
return stack;
|
|
}
|
|
};
|
|
};
|
|
|
|
/* -----[ Scope and mangling ]----- */
|
|
|
|
function Scope(parent) {
|
|
this.names = {}; // names defined in this scope
|
|
this.mangled = {}; // mangled names (orig.name => mangled)
|
|
this.rev_mangled = {}; // reverse lookup (mangled => orig.name)
|
|
this.cname = -1; // current mangled name
|
|
this.refs = {}; // names referenced from this scope
|
|
this.uses_with = false; // will become TRUE if with() is detected in this or any subscopes
|
|
this.uses_eval = false; // will become TRUE if eval() is detected in this or any subscopes
|
|
this.parent = parent; // parent scope
|
|
this.children = []; // sub-scopes
|
|
if (parent) {
|
|
this.level = parent.level + 1;
|
|
parent.children.push(this);
|
|
} else {
|
|
this.level = 0;
|
|
}
|
|
};
|
|
|
|
var base54 = (function(){
|
|
var DIGITS = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ$_";
|
|
return function(num) {
|
|
var ret = "";
|
|
do {
|
|
ret = DIGITS.charAt(num % 54) + ret;
|
|
num = Math.floor(num / 54);
|
|
} while (num > 0);
|
|
return ret;
|
|
};
|
|
})();
|
|
|
|
Scope.prototype = {
|
|
has: function(name) {
|
|
for (var s = this; s; s = s.parent)
|
|
if (HOP(s.names, name))
|
|
return s;
|
|
},
|
|
has_mangled: function(mname) {
|
|
for (var s = this; s; s = s.parent)
|
|
if (HOP(s.rev_mangled, mname))
|
|
return s;
|
|
},
|
|
toJSON: function() {
|
|
return {
|
|
names: this.names,
|
|
uses_eval: this.uses_eval,
|
|
uses_with: this.uses_with
|
|
};
|
|
},
|
|
|
|
next_mangled: function() {
|
|
// we must be careful that the new mangled name:
|
|
//
|
|
// 1. doesn't shadow a mangled name from a parent
|
|
// scope, unless we don't reference the original
|
|
// name from this scope OR from any sub-scopes!
|
|
// This will get slow.
|
|
//
|
|
// 2. doesn't shadow an original name from a parent
|
|
// scope, in the event that the name is not mangled
|
|
// in the parent scope and we reference that name
|
|
// here OR IN ANY SUBSCOPES!
|
|
//
|
|
// 3. doesn't shadow a name that is referenced but not
|
|
// defined (possibly global defined elsewhere).
|
|
for (;;) {
|
|
var m = base54(++this.cname), prior;
|
|
|
|
// case 1.
|
|
prior = this.has_mangled(m);
|
|
if (prior && this.refs[prior.rev_mangled[m]] === prior)
|
|
continue;
|
|
|
|
// case 2.
|
|
prior = this.has(m);
|
|
if (prior && prior !== this && this.refs[m] === prior && !prior.has_mangled(m))
|
|
continue;
|
|
|
|
// case 3.
|
|
if (HOP(this.refs, m) && this.refs[m] == null)
|
|
continue;
|
|
|
|
// I got "do" once. :-/
|
|
if (!is_identifier(m))
|
|
continue;
|
|
|
|
return m;
|
|
}
|
|
},
|
|
set_mangle: function(name, m) {
|
|
this.rev_mangled[m] = name;
|
|
return this.mangled[name] = m;
|
|
},
|
|
get_mangled: function(name, newMangle) {
|
|
if (this.uses_eval || this.uses_with) return name; // no mangle if eval or with is in use
|
|
var s = this.has(name);
|
|
if (!s) return name; // not in visible scope, no mangle
|
|
if (HOP(s.mangled, name)) return s.mangled[name]; // already mangled in this scope
|
|
if (!newMangle) return name; // not found and no mangling requested
|
|
return s.set_mangle(name, s.next_mangled());
|
|
},
|
|
references: function(name) {
|
|
return name && !this.parent || this.uses_with || this.uses_eval || this.refs[name];
|
|
},
|
|
define: function(name, type) {
|
|
if (name != null) {
|
|
if (type == "var" || !HOP(this.names, name))
|
|
this.names[name] = type || "var";
|
|
return name;
|
|
}
|
|
}
|
|
};
|
|
|
|
function ast_add_scope(ast) {
|
|
|
|
var current_scope = null;
|
|
var w = ast_walker(), walk = w.walk;
|
|
var having_eval = [];
|
|
|
|
function with_new_scope(cont) {
|
|
current_scope = new Scope(current_scope);
|
|
current_scope.labels = new Scope();
|
|
var ret = current_scope.body = cont();
|
|
ret.scope = current_scope;
|
|
current_scope = current_scope.parent;
|
|
return ret;
|
|
};
|
|
|
|
function define(name, type) {
|
|
return current_scope.define(name, type);
|
|
};
|
|
|
|
function reference(name) {
|
|
current_scope.refs[name] = true;
|
|
};
|
|
|
|
function _lambda(name, args, body) {
|
|
var is_defun = this[0] == "defun";
|
|
return [ this[0], is_defun ? define(name, "defun") : name, args, with_new_scope(function(){
|
|
if (!is_defun) define(name, "lambda");
|
|
MAP(args, function(name){ define(name, "arg") });
|
|
return MAP(body, walk);
|
|
})];
|
|
};
|
|
|
|
function _vardefs(type) {
|
|
return function(defs) {
|
|
MAP(defs, function(d){
|
|
define(d[0], type);
|
|
if (d[1]) reference(d[0]);
|
|
});
|
|
};
|
|
};
|
|
|
|
function _breacont(label) {
|
|
if (label)
|
|
current_scope.labels.refs[label] = true;
|
|
};
|
|
|
|
return with_new_scope(function(){
|
|
// process AST
|
|
var ret = w.with_walkers({
|
|
"function": _lambda,
|
|
"defun": _lambda,
|
|
"label": function(name, stat) { current_scope.labels.define(name) },
|
|
"break": _breacont,
|
|
"continue": _breacont,
|
|
"with": function(expr, block) {
|
|
for (var s = current_scope; s; s = s.parent)
|
|
s.uses_with = true;
|
|
},
|
|
"var": _vardefs("var"),
|
|
"const": _vardefs("const"),
|
|
"try": function(t, c, f) {
|
|
if (c != null) return [
|
|
this[0],
|
|
MAP(t, walk),
|
|
[ define(c[0], "catch"), MAP(c[1], walk) ],
|
|
f != null ? MAP(f, walk) : null
|
|
];
|
|
},
|
|
"name": function(name) {
|
|
if (name == "eval")
|
|
having_eval.push(current_scope);
|
|
reference(name);
|
|
}
|
|
}, function(){
|
|
return walk(ast);
|
|
});
|
|
|
|
// the reason why we need an additional pass here is
|
|
// that names can be used prior to their definition.
|
|
|
|
// scopes where eval was detected and their parents
|
|
// are marked with uses_eval, unless they define the
|
|
// "eval" name.
|
|
MAP(having_eval, function(scope){
|
|
if (!scope.has("eval")) while (scope) {
|
|
scope.uses_eval = true;
|
|
scope = scope.parent;
|
|
}
|
|
});
|
|
|
|
// for referenced names it might be useful to know
|
|
// their origin scope. current_scope here is the
|
|
// toplevel one.
|
|
function fixrefs(scope, i) {
|
|
// do children first; order shouldn't matter
|
|
for (i = scope.children.length; --i >= 0;)
|
|
fixrefs(scope.children[i]);
|
|
for (i in scope.refs) if (HOP(scope.refs, i)) {
|
|
// find origin scope and propagate the reference to origin
|
|
for (var origin = scope.has(i), s = scope; s; s = s.parent) {
|
|
s.refs[i] = origin;
|
|
if (s === origin) break;
|
|
}
|
|
}
|
|
};
|
|
fixrefs(current_scope);
|
|
|
|
return ret;
|
|
});
|
|
|
|
};
|
|
|
|
/* -----[ mangle names ]----- */
|
|
|
|
function ast_mangle(ast, options) {
|
|
var w = ast_walker(), walk = w.walk, scope;
|
|
options = options || {};
|
|
|
|
function get_mangled(name, newMangle) {
|
|
if (!options.toplevel && !scope.parent) return name; // don't mangle toplevel
|
|
if (options.except && member(name, options.except))
|
|
return name;
|
|
return scope.get_mangled(name, newMangle);
|
|
};
|
|
|
|
function get_define(name) {
|
|
if (options.defines) {
|
|
// we always lookup a defined symbol for the current scope FIRST, so declared
|
|
// vars trump a DEFINE symbol, but if no such var is found, then match a DEFINE value
|
|
if (!scope.has(name)) {
|
|
if (HOP(options.defines, name)) {
|
|
return options.defines[name];
|
|
}
|
|
}
|
|
return null;
|
|
}
|
|
};
|
|
|
|
function _lambda(name, args, body) {
|
|
if (!options.no_functions) {
|
|
var is_defun = this[0] == "defun", extra;
|
|
if (name) {
|
|
if (is_defun) name = get_mangled(name);
|
|
else if (body.scope.references(name)) {
|
|
extra = {};
|
|
if (!(scope.uses_eval || scope.uses_with))
|
|
name = extra[name] = scope.next_mangled();
|
|
else
|
|
extra[name] = name;
|
|
}
|
|
else name = null;
|
|
}
|
|
}
|
|
body = with_scope(body.scope, function(){
|
|
args = MAP(args, function(name){ return get_mangled(name) });
|
|
return MAP(body, walk);
|
|
}, extra);
|
|
return [ this[0], name, args, body ];
|
|
};
|
|
|
|
function with_scope(s, cont, extra) {
|
|
var _scope = scope;
|
|
scope = s;
|
|
if (extra) for (var i in extra) if (HOP(extra, i)) {
|
|
s.set_mangle(i, extra[i]);
|
|
}
|
|
for (var i in s.names) if (HOP(s.names, i)) {
|
|
get_mangled(i, true);
|
|
}
|
|
var ret = cont();
|
|
ret.scope = s;
|
|
scope = _scope;
|
|
return ret;
|
|
};
|
|
|
|
function _vardefs(defs) {
|
|
return [ this[0], MAP(defs, function(d){
|
|
return [ get_mangled(d[0]), walk(d[1]) ];
|
|
}) ];
|
|
};
|
|
|
|
function _breacont(label) {
|
|
if (label) return [ this[0], scope.labels.get_mangled(label) ];
|
|
};
|
|
|
|
return w.with_walkers({
|
|
"function": _lambda,
|
|
"defun": function() {
|
|
// move function declarations to the top when
|
|
// they are not in some block.
|
|
var ast = _lambda.apply(this, arguments);
|
|
switch (w.parent()[0]) {
|
|
case "toplevel":
|
|
case "function":
|
|
case "defun":
|
|
return MAP.at_top(ast);
|
|
}
|
|
return ast;
|
|
},
|
|
"label": function(label, stat) {
|
|
if (scope.labels.refs[label]) return [
|
|
this[0],
|
|
scope.labels.get_mangled(label, true),
|
|
walk(stat)
|
|
];
|
|
return walk(stat);
|
|
},
|
|
"break": _breacont,
|
|
"continue": _breacont,
|
|
"var": _vardefs,
|
|
"const": _vardefs,
|
|
"name": function(name) {
|
|
return get_define(name) || [ this[0], get_mangled(name) ];
|
|
},
|
|
"try": function(t, c, f) {
|
|
return [ this[0],
|
|
MAP(t, walk),
|
|
c != null ? [ get_mangled(c[0]), MAP(c[1], walk) ] : null,
|
|
f != null ? MAP(f, walk) : null ];
|
|
},
|
|
"toplevel": function(body) {
|
|
var self = this;
|
|
return with_scope(self.scope, function(){
|
|
return [ self[0], MAP(body, walk) ];
|
|
});
|
|
}
|
|
}, function() {
|
|
return walk(ast_add_scope(ast));
|
|
});
|
|
};
|
|
|
|
/* -----[
|
|
- compress foo["bar"] into foo.bar,
|
|
- remove block brackets {} where possible
|
|
- join consecutive var declarations
|
|
- various optimizations for IFs:
|
|
- if (cond) foo(); else bar(); ==> cond?foo():bar();
|
|
- if (cond) foo(); ==> cond&&foo();
|
|
- if (foo) return bar(); else return baz(); ==> return foo?bar():baz(); // also for throw
|
|
- if (foo) return bar(); else something(); ==> {if(foo)return bar();something()}
|
|
]----- */
|
|
|
|
var warn = function(){};
|
|
|
|
function best_of(ast1, ast2) {
|
|
return gen_code(ast1).length > gen_code(ast2[0] == "stat" ? ast2[1] : ast2).length ? ast2 : ast1;
|
|
};
|
|
|
|
function last_stat(b) {
|
|
if (b[0] == "block" && b[1] && b[1].length > 0)
|
|
return b[1][b[1].length - 1];
|
|
return b;
|
|
}
|
|
|
|
function aborts(t) {
|
|
if (t) switch (last_stat(t)[0]) {
|
|
case "return":
|
|
case "break":
|
|
case "continue":
|
|
case "throw":
|
|
return true;
|
|
}
|
|
};
|
|
|
|
function boolean_expr(expr) {
|
|
return ( (expr[0] == "unary-prefix"
|
|
&& member(expr[1], [ "!", "delete" ])) ||
|
|
|
|
(expr[0] == "binary"
|
|
&& member(expr[1], [ "in", "instanceof", "==", "!=", "===", "!==", "<", "<=", ">=", ">" ])) ||
|
|
|
|
(expr[0] == "binary"
|
|
&& member(expr[1], [ "&&", "||" ])
|
|
&& boolean_expr(expr[2])
|
|
&& boolean_expr(expr[3])) ||
|
|
|
|
(expr[0] == "conditional"
|
|
&& boolean_expr(expr[2])
|
|
&& boolean_expr(expr[3])) ||
|
|
|
|
(expr[0] == "assign"
|
|
&& expr[1] === true
|
|
&& boolean_expr(expr[3])) ||
|
|
|
|
(expr[0] == "seq"
|
|
&& boolean_expr(expr[expr.length - 1]))
|
|
);
|
|
};
|
|
|
|
function empty(b) {
|
|
return !b || (b[0] == "block" && (!b[1] || b[1].length == 0));
|
|
};
|
|
|
|
function is_string(node) {
|
|
return (node[0] == "string" ||
|
|
node[0] == "unary-prefix" && node[1] == "typeof" ||
|
|
node[0] == "binary" && node[1] == "+" &&
|
|
(is_string(node[2]) || is_string(node[3])));
|
|
};
|
|
|
|
var when_constant = (function(){
|
|
|
|
var $NOT_CONSTANT = {};
|
|
|
|
// this can only evaluate constant expressions. If it finds anything
|
|
// not constant, it throws $NOT_CONSTANT.
|
|
function evaluate(expr) {
|
|
switch (expr[0]) {
|
|
case "string":
|
|
case "num":
|
|
return expr[1];
|
|
case "name":
|
|
case "atom":
|
|
switch (expr[1]) {
|
|
case "true": return true;
|
|
case "false": return false;
|
|
case "null": return null;
|
|
}
|
|
break;
|
|
case "unary-prefix":
|
|
switch (expr[1]) {
|
|
case "!": return !evaluate(expr[2]);
|
|
case "typeof": return typeof evaluate(expr[2]);
|
|
case "~": return ~evaluate(expr[2]);
|
|
case "-": return -evaluate(expr[2]);
|
|
case "+": return +evaluate(expr[2]);
|
|
}
|
|
break;
|
|
case "binary":
|
|
var left = expr[2], right = expr[3];
|
|
switch (expr[1]) {
|
|
case "&&" : return evaluate(left) && evaluate(right);
|
|
case "||" : return evaluate(left) || evaluate(right);
|
|
case "|" : return evaluate(left) | evaluate(right);
|
|
case "&" : return evaluate(left) & evaluate(right);
|
|
case "^" : return evaluate(left) ^ evaluate(right);
|
|
case "+" : return evaluate(left) + evaluate(right);
|
|
case "*" : return evaluate(left) * evaluate(right);
|
|
case "/" : return evaluate(left) / evaluate(right);
|
|
case "%" : return evaluate(left) % evaluate(right);
|
|
case "-" : return evaluate(left) - evaluate(right);
|
|
case "<<" : return evaluate(left) << evaluate(right);
|
|
case ">>" : return evaluate(left) >> evaluate(right);
|
|
case ">>>" : return evaluate(left) >>> evaluate(right);
|
|
case "==" : return evaluate(left) == evaluate(right);
|
|
case "===" : return evaluate(left) === evaluate(right);
|
|
case "!=" : return evaluate(left) != evaluate(right);
|
|
case "!==" : return evaluate(left) !== evaluate(right);
|
|
case "<" : return evaluate(left) < evaluate(right);
|
|
case "<=" : return evaluate(left) <= evaluate(right);
|
|
case ">" : return evaluate(left) > evaluate(right);
|
|
case ">=" : return evaluate(left) >= evaluate(right);
|
|
case "in" : return evaluate(left) in evaluate(right);
|
|
case "instanceof" : return evaluate(left) instanceof evaluate(right);
|
|
}
|
|
}
|
|
throw $NOT_CONSTANT;
|
|
};
|
|
|
|
return function(expr, yes, no) {
|
|
try {
|
|
var val = evaluate(expr), ast;
|
|
switch (typeof val) {
|
|
case "string": ast = [ "string", val ]; break;
|
|
case "number": ast = [ "num", val ]; break;
|
|
case "boolean": ast = [ "name", String(val) ]; break;
|
|
default:
|
|
if (val === null) { ast = [ "atom", "null" ]; break; }
|
|
throw new Error("Can't handle constant of type: " + (typeof val));
|
|
}
|
|
return yes.call(expr, ast, val);
|
|
} catch(ex) {
|
|
if (ex === $NOT_CONSTANT) {
|
|
if (expr[0] == "binary"
|
|
&& (expr[1] == "===" || expr[1] == "!==")
|
|
&& ((is_string(expr[2]) && is_string(expr[3]))
|
|
|| (boolean_expr(expr[2]) && boolean_expr(expr[3])))) {
|
|
expr[1] = expr[1].substr(0, 2);
|
|
}
|
|
else if (no && expr[0] == "binary"
|
|
&& (expr[1] == "||" || expr[1] == "&&")) {
|
|
// the whole expression is not constant but the lval may be...
|
|
try {
|
|
var lval = evaluate(expr[2]);
|
|
expr = ((expr[1] == "&&" && (lval ? expr[3] : lval)) ||
|
|
(expr[1] == "||" && (lval ? lval : expr[3])) ||
|
|
expr);
|
|
} catch(ex2) {
|
|
// IGNORE... lval is not constant
|
|
}
|
|
}
|
|
return no ? no.call(expr, expr) : null;
|
|
}
|
|
else throw ex;
|
|
}
|
|
};
|
|
|
|
})();
|
|
|
|
function warn_unreachable(ast) {
|
|
if (!empty(ast))
|
|
warn("Dropping unreachable code: " + gen_code(ast, true));
|
|
};
|
|
|
|
function prepare_ifs(ast) {
|
|
var w = ast_walker(), walk = w.walk;
|
|
// In this first pass, we rewrite ifs which abort with no else with an
|
|
// if-else. For example:
|
|
//
|
|
// if (x) {
|
|
// blah();
|
|
// return y;
|
|
// }
|
|
// foobar();
|
|
//
|
|
// is rewritten into:
|
|
//
|
|
// if (x) {
|
|
// blah();
|
|
// return y;
|
|
// } else {
|
|
// foobar();
|
|
// }
|
|
function redo_if(statements) {
|
|
statements = MAP(statements, walk);
|
|
|
|
for (var i = 0; i < statements.length; ++i) {
|
|
var fi = statements[i];
|
|
if (fi[0] != "if") continue;
|
|
|
|
if (fi[3] && walk(fi[3])) continue;
|
|
|
|
var t = walk(fi[2]);
|
|
if (!aborts(t)) continue;
|
|
|
|
var conditional = walk(fi[1]);
|
|
|
|
var e_body = redo_if(statements.slice(i + 1));
|
|
var e = e_body.length == 1 ? e_body[0] : [ "block", e_body ];
|
|
|
|
return statements.slice(0, i).concat([ [
|
|
fi[0], // "if"
|
|
conditional, // conditional
|
|
t, // then
|
|
e // else
|
|
] ]);
|
|
}
|
|
|
|
return statements;
|
|
};
|
|
|
|
function redo_if_lambda(name, args, body) {
|
|
body = redo_if(body);
|
|
return [ this[0], name, args, body ];
|
|
};
|
|
|
|
function redo_if_block(statements) {
|
|
return [ this[0], statements != null ? redo_if(statements) : null ];
|
|
};
|
|
|
|
return w.with_walkers({
|
|
"defun": redo_if_lambda,
|
|
"function": redo_if_lambda,
|
|
"block": redo_if_block,
|
|
"splice": redo_if_block,
|
|
"toplevel": function(statements) {
|
|
return [ this[0], redo_if(statements) ];
|
|
},
|
|
"try": function(t, c, f) {
|
|
return [
|
|
this[0],
|
|
redo_if(t),
|
|
c != null ? [ c[0], redo_if(c[1]) ] : null,
|
|
f != null ? redo_if(f) : null
|
|
];
|
|
}
|
|
}, function() {
|
|
return walk(ast);
|
|
});
|
|
};
|
|
|
|
function for_side_effects(ast, handler) {
|
|
var w = ast_walker(), walk = w.walk;
|
|
var $stop = {}, $restart = {};
|
|
function stop() { throw $stop };
|
|
function restart() { throw $restart };
|
|
function found(){ return handler.call(this, this, w, stop, restart) };
|
|
function unary(op) {
|
|
if (op == "++" || op == "--")
|
|
return found.apply(this, arguments);
|
|
};
|
|
return w.with_walkers({
|
|
"try": found,
|
|
"throw": found,
|
|
"return": found,
|
|
"new": found,
|
|
"switch": found,
|
|
"break": found,
|
|
"continue": found,
|
|
"assign": found,
|
|
"call": found,
|
|
"if": found,
|
|
"for": found,
|
|
"for-in": found,
|
|
"while": found,
|
|
"do": found,
|
|
"return": found,
|
|
"unary-prefix": unary,
|
|
"unary-postfix": unary,
|
|
"defun": found
|
|
}, function(){
|
|
while (true) try {
|
|
walk(ast);
|
|
break;
|
|
} catch(ex) {
|
|
if (ex === $stop) break;
|
|
if (ex === $restart) continue;
|
|
throw ex;
|
|
}
|
|
});
|
|
};
|
|
|
|
function ast_lift_variables(ast) {
|
|
var w = ast_walker(), walk = w.walk, scope;
|
|
function do_body(body, env) {
|
|
var _scope = scope;
|
|
scope = env;
|
|
body = MAP(body, walk);
|
|
var hash = {}, names = MAP(env.names, function(type, name){
|
|
if (type != "var") return MAP.skip;
|
|
if (!env.references(name)) return MAP.skip;
|
|
hash[name] = true;
|
|
return [ name ];
|
|
});
|
|
if (names.length > 0) {
|
|
// looking for assignments to any of these variables.
|
|
// we can save considerable space by moving the definitions
|
|
// in the var declaration.
|
|
for_side_effects([ "block", body ], function(ast, walker, stop, restart) {
|
|
if (ast[0] == "assign"
|
|
&& ast[1] === true
|
|
&& ast[2][0] == "name"
|
|
&& HOP(hash, ast[2][1])) {
|
|
// insert the definition into the var declaration
|
|
for (var i = names.length; --i >= 0;) {
|
|
if (names[i][0] == ast[2][1]) {
|
|
if (names[i][1]) // this name already defined, we must stop
|
|
stop();
|
|
names[i][1] = ast[3]; // definition
|
|
names.push(names.splice(i, 1)[0]);
|
|
break;
|
|
}
|
|
}
|
|
// remove this assignment from the AST.
|
|
var p = walker.parent();
|
|
if (p[0] == "seq") {
|
|
var a = p[2];
|
|
a.unshift(0, p.length);
|
|
p.splice.apply(p, a);
|
|
}
|
|
else if (p[0] == "stat") {
|
|
p.splice(0, p.length, "block"); // empty statement
|
|
}
|
|
else {
|
|
stop();
|
|
}
|
|
restart();
|
|
}
|
|
stop();
|
|
});
|
|
body.unshift([ "var", names ]);
|
|
}
|
|
scope = _scope;
|
|
return body;
|
|
};
|
|
function _vardefs(defs) {
|
|
var ret = null;
|
|
for (var i = defs.length; --i >= 0;) {
|
|
var d = defs[i];
|
|
if (!d[1]) continue;
|
|
d = [ "assign", true, [ "name", d[0] ], d[1] ];
|
|
if (ret == null) ret = d;
|
|
else ret = [ "seq", d, ret ];
|
|
}
|
|
if (ret == null) {
|
|
if (w.parent()[0] == "for-in")
|
|
return [ "name", defs[0][0] ];
|
|
return MAP.skip;
|
|
}
|
|
return [ "stat", ret ];
|
|
};
|
|
function _toplevel(body) {
|
|
return [ this[0], do_body(body, this.scope) ];
|
|
};
|
|
return w.with_walkers({
|
|
"function": function(name, args, body){
|
|
for (var i = args.length; --i >= 0 && !body.scope.references(args[i]);)
|
|
args.pop();
|
|
if (!body.scope.references(name)) name = null;
|
|
return [ this[0], name, args, do_body(body, body.scope) ];
|
|
},
|
|
"defun": function(name, args, body){
|
|
if (!scope.references(name)) return MAP.skip;
|
|
for (var i = args.length; --i >= 0 && !body.scope.references(args[i]);)
|
|
args.pop();
|
|
return [ this[0], name, args, do_body(body, body.scope) ];
|
|
},
|
|
"var": _vardefs,
|
|
"toplevel": _toplevel
|
|
}, function(){
|
|
return walk(ast_add_scope(ast));
|
|
});
|
|
};
|
|
|
|
function ast_squeeze(ast, options) {
|
|
options = defaults(options, {
|
|
make_seqs : true,
|
|
dead_code : true,
|
|
no_warnings : false,
|
|
keep_comps : true
|
|
});
|
|
|
|
var w = ast_walker(), walk = w.walk;
|
|
|
|
function negate(c) {
|
|
var not_c = [ "unary-prefix", "!", c ];
|
|
switch (c[0]) {
|
|
case "unary-prefix":
|
|
return c[1] == "!" && boolean_expr(c[2]) ? c[2] : not_c;
|
|
case "seq":
|
|
c = slice(c);
|
|
c[c.length - 1] = negate(c[c.length - 1]);
|
|
return c;
|
|
case "conditional":
|
|
return best_of(not_c, [ "conditional", c[1], negate(c[2]), negate(c[3]) ]);
|
|
case "binary":
|
|
var op = c[1], left = c[2], right = c[3];
|
|
if (!options.keep_comps) switch (op) {
|
|
case "<=" : return [ "binary", ">", left, right ];
|
|
case "<" : return [ "binary", ">=", left, right ];
|
|
case ">=" : return [ "binary", "<", left, right ];
|
|
case ">" : return [ "binary", "<=", left, right ];
|
|
}
|
|
switch (op) {
|
|
case "==" : return [ "binary", "!=", left, right ];
|
|
case "!=" : return [ "binary", "==", left, right ];
|
|
case "===" : return [ "binary", "!==", left, right ];
|
|
case "!==" : return [ "binary", "===", left, right ];
|
|
case "&&" : return best_of(not_c, [ "binary", "||", negate(left), negate(right) ]);
|
|
case "||" : return best_of(not_c, [ "binary", "&&", negate(left), negate(right) ]);
|
|
}
|
|
break;
|
|
}
|
|
return not_c;
|
|
};
|
|
|
|
function make_conditional(c, t, e) {
|
|
var make_real_conditional = function() {
|
|
if (c[0] == "unary-prefix" && c[1] == "!") {
|
|
return e ? [ "conditional", c[2], e, t ] : [ "binary", "||", c[2], t ];
|
|
} else {
|
|
return e ? best_of(
|
|
[ "conditional", c, t, e ],
|
|
[ "conditional", negate(c), e, t ]
|
|
) : [ "binary", "&&", c, t ];
|
|
}
|
|
};
|
|
// shortcut the conditional if the expression has a constant value
|
|
return when_constant(c, function(ast, val){
|
|
warn_unreachable(val ? e : t);
|
|
return (val ? t : e);
|
|
}, make_real_conditional);
|
|
};
|
|
|
|
function rmblock(block) {
|
|
if (block != null && block[0] == "block" && block[1]) {
|
|
if (block[1].length == 1)
|
|
block = block[1][0];
|
|
else if (block[1].length == 0)
|
|
block = [ "block" ];
|
|
}
|
|
return block;
|
|
};
|
|
|
|
function _lambda(name, args, body) {
|
|
return [ this[0], name, args, tighten(body, "lambda") ];
|
|
};
|
|
|
|
// this function does a few things:
|
|
// 1. discard useless blocks
|
|
// 2. join consecutive var declarations
|
|
// 3. remove obviously dead code
|
|
// 4. transform consecutive statements using the comma operator
|
|
// 5. if block_type == "lambda" and it detects constructs like if(foo) return ... - rewrite like if (!foo) { ... }
|
|
function tighten(statements, block_type) {
|
|
statements = MAP(statements, walk);
|
|
|
|
statements = statements.reduce(function(a, stat){
|
|
if (stat[0] == "block") {
|
|
if (stat[1]) {
|
|
a.push.apply(a, stat[1]);
|
|
}
|
|
} else {
|
|
a.push(stat);
|
|
}
|
|
return a;
|
|
}, []);
|
|
|
|
statements = (function(a, prev){
|
|
statements.forEach(function(cur){
|
|
if (prev && ((cur[0] == "var" && prev[0] == "var") ||
|
|
(cur[0] == "const" && prev[0] == "const"))) {
|
|
prev[1] = prev[1].concat(cur[1]);
|
|
} else {
|
|
a.push(cur);
|
|
prev = cur;
|
|
}
|
|
});
|
|
return a;
|
|
})([]);
|
|
|
|
if (options.dead_code) statements = (function(a, has_quit){
|
|
statements.forEach(function(st){
|
|
if (has_quit) {
|
|
if (st[0] == "function" || st[0] == "defun") {
|
|
a.push(st);
|
|
}
|
|
else if (st[0] == "var" || st[0] == "const") {
|
|
if (!options.no_warnings)
|
|
warn("Variables declared in unreachable code");
|
|
st[1] = MAP(st[1], function(def){
|
|
if (def[1] && !options.no_warnings)
|
|
warn_unreachable([ "assign", true, [ "name", def[0] ], def[1] ]);
|
|
return [ def[0] ];
|
|
});
|
|
a.push(st);
|
|
}
|
|
else if (!options.no_warnings)
|
|
warn_unreachable(st);
|
|
}
|
|
else {
|
|
a.push(st);
|
|
if (member(st[0], [ "return", "throw", "break", "continue" ]))
|
|
has_quit = true;
|
|
}
|
|
});
|
|
return a;
|
|
})([]);
|
|
|
|
if (options.make_seqs) statements = (function(a, prev) {
|
|
statements.forEach(function(cur){
|
|
if (prev && prev[0] == "stat" && cur[0] == "stat") {
|
|
prev[1] = [ "seq", prev[1], cur[1] ];
|
|
} else {
|
|
a.push(cur);
|
|
prev = cur;
|
|
}
|
|
});
|
|
if (a.length >= 2
|
|
&& a[a.length-2][0] == "stat"
|
|
&& (a[a.length-1][0] == "return" || a[a.length-1][0] == "throw")
|
|
&& a[a.length-1][1])
|
|
{
|
|
a.splice(a.length - 2, 2,
|
|
[ a[a.length-1][0],
|
|
[ "seq", a[a.length-2][1], a[a.length-1][1] ]]);
|
|
}
|
|
return a;
|
|
})([]);
|
|
|
|
// this increases jQuery by 1K. Probably not such a good idea after all..
|
|
// part of this is done in prepare_ifs anyway.
|
|
// if (block_type == "lambda") statements = (function(i, a, stat){
|
|
// while (i < statements.length) {
|
|
// stat = statements[i++];
|
|
// if (stat[0] == "if" && !stat[3]) {
|
|
// if (stat[2][0] == "return" && stat[2][1] == null) {
|
|
// a.push(make_if(negate(stat[1]), [ "block", statements.slice(i) ]));
|
|
// break;
|
|
// }
|
|
// var last = last_stat(stat[2]);
|
|
// if (last[0] == "return" && last[1] == null) {
|
|
// a.push(make_if(stat[1], [ "block", stat[2][1].slice(0, -1) ], [ "block", statements.slice(i) ]));
|
|
// break;
|
|
// }
|
|
// }
|
|
// a.push(stat);
|
|
// }
|
|
// return a;
|
|
// })(0, []);
|
|
|
|
return statements;
|
|
};
|
|
|
|
function make_if(c, t, e) {
|
|
return when_constant(c, function(ast, val){
|
|
if (val) {
|
|
t = walk(t);
|
|
warn_unreachable(e);
|
|
return t || [ "block" ];
|
|
} else {
|
|
e = walk(e);
|
|
warn_unreachable(t);
|
|
return e || [ "block" ];
|
|
}
|
|
}, function() {
|
|
return make_real_if(c, t, e);
|
|
});
|
|
};
|
|
|
|
function abort_else(c, t, e) {
|
|
var ret = [ [ "if", negate(c), e ] ];
|
|
if (t[0] == "block") {
|
|
if (t[1]) ret = ret.concat(t[1]);
|
|
} else {
|
|
ret.push(t);
|
|
}
|
|
return walk([ "block", ret ]);
|
|
};
|
|
|
|
function make_real_if(c, t, e) {
|
|
c = walk(c);
|
|
t = walk(t);
|
|
e = walk(e);
|
|
|
|
if (empty(t)) {
|
|
c = negate(c);
|
|
t = e;
|
|
e = null;
|
|
} else if (empty(e)) {
|
|
e = null;
|
|
} else {
|
|
// if we have both else and then, maybe it makes sense to switch them?
|
|
(function(){
|
|
var a = gen_code(c);
|
|
var n = negate(c);
|
|
var b = gen_code(n);
|
|
if (b.length < a.length) {
|
|
var tmp = t;
|
|
t = e;
|
|
e = tmp;
|
|
c = n;
|
|
}
|
|
})();
|
|
}
|
|
if (empty(e) && empty(t))
|
|
return [ "stat", c ];
|
|
var ret = [ "if", c, t, e ];
|
|
if (t[0] == "if" && empty(t[3]) && empty(e)) {
|
|
ret = best_of(ret, walk([ "if", [ "binary", "&&", c, t[1] ], t[2] ]));
|
|
}
|
|
else if (t[0] == "stat") {
|
|
if (e) {
|
|
if (e[0] == "stat")
|
|
ret = best_of(ret, [ "stat", make_conditional(c, t[1], e[1]) ]);
|
|
else if (aborts(e))
|
|
ret = abort_else(c, t, e);
|
|
}
|
|
else {
|
|
ret = best_of(ret, [ "stat", make_conditional(c, t[1]) ]);
|
|
}
|
|
}
|
|
else if (e && t[0] == e[0] && (t[0] == "return" || t[0] == "throw") && t[1] && e[1]) {
|
|
ret = best_of(ret, [ t[0], make_conditional(c, t[1], e[1] ) ]);
|
|
}
|
|
else if (e && aborts(t)) {
|
|
ret = [ [ "if", c, t ] ];
|
|
if (e[0] == "block") {
|
|
if (e[1]) ret = ret.concat(e[1]);
|
|
}
|
|
else {
|
|
ret.push(e);
|
|
}
|
|
ret = walk([ "block", ret ]);
|
|
}
|
|
else if (t && aborts(e)) {
|
|
ret = abort_else(c, t, e);
|
|
}
|
|
return ret;
|
|
};
|
|
|
|
function _do_while(cond, body) {
|
|
return when_constant(cond, function(cond, val){
|
|
if (!val) {
|
|
warn_unreachable(body);
|
|
return [ "block" ];
|
|
} else {
|
|
return [ "for", null, null, null, walk(body) ];
|
|
}
|
|
});
|
|
};
|
|
|
|
return w.with_walkers({
|
|
"sub": function(expr, subscript) {
|
|
if (subscript[0] == "string") {
|
|
var name = subscript[1];
|
|
if (is_identifier(name))
|
|
return [ "dot", walk(expr), name ];
|
|
else if (/^[1-9][0-9]*$/.test(name) || name === "0")
|
|
return [ "sub", walk(expr), [ "num", parseInt(name, 10) ] ];
|
|
}
|
|
},
|
|
"if": make_if,
|
|
"toplevel": function(body) {
|
|
return [ "toplevel", tighten(body) ];
|
|
},
|
|
"switch": function(expr, body) {
|
|
var last = body.length - 1;
|
|
return [ "switch", walk(expr), MAP(body, function(branch, i){
|
|
var block = tighten(branch[1]);
|
|
if (i == last && block.length > 0) {
|
|
var node = block[block.length - 1];
|
|
if (node[0] == "break" && !node[1])
|
|
block.pop();
|
|
}
|
|
return [ branch[0] ? walk(branch[0]) : null, block ];
|
|
}) ];
|
|
},
|
|
"function": _lambda,
|
|
"defun": _lambda,
|
|
"block": function(body) {
|
|
if (body) return rmblock([ "block", tighten(body) ]);
|
|
},
|
|
"binary": function(op, left, right) {
|
|
return when_constant([ "binary", op, walk(left), walk(right) ], function yes(c){
|
|
return best_of(walk(c), this);
|
|
}, function no() {
|
|
return function(){
|
|
if(op != "==" && op != "!=") return;
|
|
var l = walk(left), r = walk(right);
|
|
if(l && l[0] == "unary-prefix" && l[1] == "!" && l[2][0] == "num")
|
|
left = ['num', +!l[2][1]];
|
|
else if (r && r[0] == "unary-prefix" && r[1] == "!" && r[2][0] == "num")
|
|
right = ['num', +!r[2][1]];
|
|
return ["binary", op, left, right];
|
|
}() || this;
|
|
});
|
|
},
|
|
"conditional": function(c, t, e) {
|
|
return make_conditional(walk(c), walk(t), walk(e));
|
|
},
|
|
"try": function(t, c, f) {
|
|
return [
|
|
"try",
|
|
tighten(t),
|
|
c != null ? [ c[0], tighten(c[1]) ] : null,
|
|
f != null ? tighten(f) : null
|
|
];
|
|
},
|
|
"unary-prefix": function(op, expr) {
|
|
expr = walk(expr);
|
|
var ret = [ "unary-prefix", op, expr ];
|
|
if (op == "!")
|
|
ret = best_of(ret, negate(expr));
|
|
return when_constant(ret, function(ast, val){
|
|
return walk(ast); // it's either true or false, so minifies to !0 or !1
|
|
}, function() { return ret });
|
|
},
|
|
"name": function(name) {
|
|
switch (name) {
|
|
case "true": return [ "unary-prefix", "!", [ "num", 0 ]];
|
|
case "false": return [ "unary-prefix", "!", [ "num", 1 ]];
|
|
}
|
|
},
|
|
"while": _do_while,
|
|
"assign": function(op, lvalue, rvalue) {
|
|
lvalue = walk(lvalue);
|
|
rvalue = walk(rvalue);
|
|
var okOps = [ '+', '-', '/', '*', '%', '>>', '<<', '>>>', '|', '^', '&' ];
|
|
if (op === true && lvalue[0] === "name" && rvalue[0] === "binary" &&
|
|
~okOps.indexOf(rvalue[1]) && rvalue[2][0] === "name" &&
|
|
rvalue[2][1] === lvalue[1]) {
|
|
return [ this[0], rvalue[1], lvalue, rvalue[3] ]
|
|
}
|
|
return [ this[0], op, lvalue, rvalue ];
|
|
}
|
|
}, function() {
|
|
for (var i = 0; i < 2; ++i) {
|
|
ast = prepare_ifs(ast);
|
|
ast = walk(ast);
|
|
}
|
|
return ast;
|
|
});
|
|
};
|
|
|
|
/* -----[ re-generate code from the AST ]----- */
|
|
|
|
var DOT_CALL_NO_PARENS = jsp.array_to_hash([
|
|
"name",
|
|
"array",
|
|
"object",
|
|
"string",
|
|
"dot",
|
|
"sub",
|
|
"call",
|
|
"regexp",
|
|
"defun"
|
|
]);
|
|
|
|
function make_string(str, ascii_only) {
|
|
var dq = 0, sq = 0;
|
|
str = str.replace(/[\\\b\f\n\r\t\x22\x27\u2028\u2029\0]/g, function(s){
|
|
switch (s) {
|
|
case "\\": return "\\\\";
|
|
case "\b": return "\\b";
|
|
case "\f": return "\\f";
|
|
case "\n": return "\\n";
|
|
case "\r": return "\\r";
|
|
case "\t": return "\\t";
|
|
case "\u2028": return "\\u2028";
|
|
case "\u2029": return "\\u2029";
|
|
case '"': ++dq; return '"';
|
|
case "'": ++sq; return "'";
|
|
case "\0": return "\\0";
|
|
}
|
|
return s;
|
|
});
|
|
if (ascii_only) str = to_ascii(str);
|
|
if (dq > sq) return "'" + str.replace(/\x27/g, "\\'") + "'";
|
|
else return '"' + str.replace(/\x22/g, '\\"') + '"';
|
|
};
|
|
|
|
function to_ascii(str) {
|
|
return str.replace(/[\u0080-\uffff]/g, function(ch) {
|
|
var code = ch.charCodeAt(0).toString(16);
|
|
while (code.length < 4) code = "0" + code;
|
|
return "\\u" + code;
|
|
});
|
|
};
|
|
|
|
var SPLICE_NEEDS_BRACKETS = jsp.array_to_hash([ "if", "while", "do", "for", "for-in", "with" ]);
|
|
|
|
function gen_code(ast, options) {
|
|
options = defaults(options, {
|
|
indent_start : 0,
|
|
indent_level : 4,
|
|
quote_keys : false,
|
|
space_colon : false,
|
|
beautify : false,
|
|
ascii_only : false,
|
|
inline_script: false
|
|
});
|
|
var beautify = !!options.beautify;
|
|
var indentation = 0,
|
|
newline = beautify ? "\n" : "",
|
|
space = beautify ? " " : "";
|
|
|
|
function encode_string(str) {
|
|
var ret = make_string(str, options.ascii_only);
|
|
if (options.inline_script)
|
|
ret = ret.replace(/<\x2fscript([>\/\t\n\f\r ])/gi, "<\\/script$1");
|
|
return ret;
|
|
};
|
|
|
|
function make_name(name) {
|
|
name = name.toString();
|
|
if (options.ascii_only)
|
|
name = to_ascii(name);
|
|
return name;
|
|
};
|
|
|
|
function indent(line) {
|
|
if (line == null)
|
|
line = "";
|
|
if (beautify)
|
|
line = repeat_string(" ", options.indent_start + indentation * options.indent_level) + line;
|
|
return line;
|
|
};
|
|
|
|
function with_indent(cont, incr) {
|
|
if (incr == null) incr = 1;
|
|
indentation += incr;
|
|
try { return cont.apply(null, slice(arguments, 1)); }
|
|
finally { indentation -= incr; }
|
|
};
|
|
|
|
function add_spaces(a) {
|
|
if (beautify)
|
|
return a.join(" ");
|
|
var b = [];
|
|
for (var i = 0; i < a.length; ++i) {
|
|
var next = a[i + 1];
|
|
b.push(a[i]);
|
|
if (next &&
|
|
((/[a-z0-9_\x24]$/i.test(a[i].toString()) && /^[a-z0-9_\x24]/i.test(next.toString())) ||
|
|
(/[\+\-]$/.test(a[i].toString()) && /^[\+\-]/.test(next.toString())))) {
|
|
b.push(" ");
|
|
}
|
|
}
|
|
return b.join("");
|
|
};
|
|
|
|
function add_commas(a) {
|
|
return a.join("," + space);
|
|
};
|
|
|
|
function parenthesize(expr) {
|
|
var gen = make(expr);
|
|
for (var i = 1; i < arguments.length; ++i) {
|
|
var el = arguments[i];
|
|
if ((el instanceof Function && el(expr)) || expr[0] == el)
|
|
return "(" + gen + ")";
|
|
}
|
|
return gen;
|
|
};
|
|
|
|
function best_of(a) {
|
|
if (a.length == 1) {
|
|
return a[0];
|
|
}
|
|
if (a.length == 2) {
|
|
var b = a[1];
|
|
a = a[0];
|
|
return a.length <= b.length ? a : b;
|
|
}
|
|
return best_of([ a[0], best_of(a.slice(1)) ]);
|
|
};
|
|
|
|
function needs_parens(expr) {
|
|
if (expr[0] == "function" || expr[0] == "object") {
|
|
// dot/call on a literal function requires the
|
|
// function literal itself to be parenthesized
|
|
// only if it's the first "thing" in a
|
|
// statement. This means that the parent is
|
|
// "stat", but it could also be a "seq" and
|
|
// we're the first in this "seq" and the
|
|
// parent is "stat", and so on. Messy stuff,
|
|
// but it worths the trouble.
|
|
var a = slice(w.stack()), self = a.pop(), p = a.pop();
|
|
while (p) {
|
|
if (p[0] == "stat") return true;
|
|
if (((p[0] == "seq" || p[0] == "call" || p[0] == "dot" || p[0] == "sub" || p[0] == "conditional") && p[1] === self) ||
|
|
((p[0] == "binary" || p[0] == "assign" || p[0] == "unary-postfix") && p[2] === self)) {
|
|
self = p;
|
|
p = a.pop();
|
|
} else {
|
|
return false;
|
|
}
|
|
}
|
|
}
|
|
return !HOP(DOT_CALL_NO_PARENS, expr[0]);
|
|
};
|
|
|
|
function make_num(num) {
|
|
var str = num.toString(10), a = [ str.replace(/^0\./, ".") ], m;
|
|
if (Math.floor(num) === num) {
|
|
if (num >= 0) {
|
|
a.push("0x" + num.toString(16).toLowerCase(), // probably pointless
|
|
"0" + num.toString(8)); // same.
|
|
} else {
|
|
a.push("-0x" + (-num).toString(16).toLowerCase(), // probably pointless
|
|
"-0" + (-num).toString(8)); // same.
|
|
}
|
|
if ((m = /^(.*?)(0+)$/.exec(num))) {
|
|
a.push(m[1] + "e" + m[2].length);
|
|
}
|
|
} else if ((m = /^0?\.(0+)(.*)$/.exec(num))) {
|
|
a.push(m[2] + "e-" + (m[1].length + m[2].length),
|
|
str.substr(str.indexOf(".")));
|
|
}
|
|
return best_of(a);
|
|
};
|
|
|
|
var w = ast_walker();
|
|
var make = w.walk;
|
|
return w.with_walkers({
|
|
"string": encode_string,
|
|
"num": make_num,
|
|
"name": make_name,
|
|
"debugger": function(){ return "debugger" },
|
|
"toplevel": function(statements) {
|
|
return make_block_statements(statements)
|
|
.join(newline + newline);
|
|
},
|
|
"splice": function(statements) {
|
|
var parent = w.parent();
|
|
if (HOP(SPLICE_NEEDS_BRACKETS, parent)) {
|
|
// we need block brackets in this case
|
|
return make_block.apply(this, arguments);
|
|
} else {
|
|
return MAP(make_block_statements(statements, true),
|
|
function(line, i) {
|
|
// the first line is already indented
|
|
return i > 0 ? indent(line) : line;
|
|
}).join(newline);
|
|
}
|
|
},
|
|
"block": make_block,
|
|
"var": function(defs) {
|
|
return "var " + add_commas(MAP(defs, make_1vardef)) + ";";
|
|
},
|
|
"const": function(defs) {
|
|
return "const " + add_commas(MAP(defs, make_1vardef)) + ";";
|
|
},
|
|
"try": function(tr, ca, fi) {
|
|
var out = [ "try", make_block(tr) ];
|
|
if (ca) out.push("catch", "(" + ca[0] + ")", make_block(ca[1]));
|
|
if (fi) out.push("finally", make_block(fi));
|
|
return add_spaces(out);
|
|
},
|
|
"throw": function(expr) {
|
|
return add_spaces([ "throw", make(expr) ]) + ";";
|
|
},
|
|
"new": function(ctor, args) {
|
|
args = args.length > 0 ? "(" + add_commas(MAP(args, function(expr){
|
|
return parenthesize(expr, "seq");
|
|
})) + ")" : "";
|
|
return add_spaces([ "new", parenthesize(ctor, "seq", "binary", "conditional", "assign", function(expr){
|
|
var w = ast_walker(), has_call = {};
|
|
try {
|
|
w.with_walkers({
|
|
"call": function() { throw has_call },
|
|
"function": function() { return this }
|
|
}, function(){
|
|
w.walk(expr);
|
|
});
|
|
} catch(ex) {
|
|
if (ex === has_call)
|
|
return true;
|
|
throw ex;
|
|
}
|
|
}) + args ]);
|
|
},
|
|
"switch": function(expr, body) {
|
|
return add_spaces([ "switch", "(" + make(expr) + ")", make_switch_block(body) ]);
|
|
},
|
|
"break": function(label) {
|
|
var out = "break";
|
|
if (label != null)
|
|
out += " " + make_name(label);
|
|
return out + ";";
|
|
},
|
|
"continue": function(label) {
|
|
var out = "continue";
|
|
if (label != null)
|
|
out += " " + make_name(label);
|
|
return out + ";";
|
|
},
|
|
"conditional": function(co, th, el) {
|
|
return add_spaces([ parenthesize(co, "assign", "seq", "conditional"), "?",
|
|
parenthesize(th, "seq"), ":",
|
|
parenthesize(el, "seq") ]);
|
|
},
|
|
"assign": function(op, lvalue, rvalue) {
|
|
if (op && op !== true) op += "=";
|
|
else op = "=";
|
|
return add_spaces([ make(lvalue), op, parenthesize(rvalue, "seq") ]);
|
|
},
|
|
"dot": function(expr) {
|
|
var out = make(expr), i = 1;
|
|
if (expr[0] == "num") {
|
|
if (!/\./.test(expr[1]))
|
|
out += ".";
|
|
} else if (needs_parens(expr))
|
|
out = "(" + out + ")";
|
|
while (i < arguments.length)
|
|
out += "." + make_name(arguments[i++]);
|
|
return out;
|
|
},
|
|
"call": function(func, args) {
|
|
var f = make(func);
|
|
if (f.charAt(0) != "(" && needs_parens(func))
|
|
f = "(" + f + ")";
|
|
return f + "(" + add_commas(MAP(args, function(expr){
|
|
return parenthesize(expr, "seq");
|
|
})) + ")";
|
|
},
|
|
"function": make_function,
|
|
"defun": make_function,
|
|
"if": function(co, th, el) {
|
|
var out = [ "if", "(" + make(co) + ")", el ? make_then(th) : make(th) ];
|
|
if (el) {
|
|
out.push("else", make(el));
|
|
}
|
|
return add_spaces(out);
|
|
},
|
|
"for": function(init, cond, step, block) {
|
|
var out = [ "for" ];
|
|
init = (init != null ? make(init) : "").replace(/;*\s*$/, ";" + space);
|
|
cond = (cond != null ? make(cond) : "").replace(/;*\s*$/, ";" + space);
|
|
step = (step != null ? make(step) : "").replace(/;*\s*$/, "");
|
|
var args = init + cond + step;
|
|
if (args == "; ; ") args = ";;";
|
|
out.push("(" + args + ")", make(block));
|
|
return add_spaces(out);
|
|
},
|
|
"for-in": function(vvar, key, hash, block) {
|
|
return add_spaces([ "for", "(" +
|
|
(vvar ? make(vvar).replace(/;+$/, "") : make(key)),
|
|
"in",
|
|
make(hash) + ")", make(block) ]);
|
|
},
|
|
"while": function(condition, block) {
|
|
return add_spaces([ "while", "(" + make(condition) + ")", make(block) ]);
|
|
},
|
|
"do": function(condition, block) {
|
|
return add_spaces([ "do", make(block), "while", "(" + make(condition) + ")" ]) + ";";
|
|
},
|
|
"return": function(expr) {
|
|
var out = [ "return" ];
|
|
if (expr != null) out.push(make(expr));
|
|
return add_spaces(out) + ";";
|
|
},
|
|
"binary": function(operator, lvalue, rvalue) {
|
|
var left = make(lvalue), right = make(rvalue);
|
|
// XXX: I'm pretty sure other cases will bite here.
|
|
// we need to be smarter.
|
|
// adding parens all the time is the safest bet.
|
|
if (member(lvalue[0], [ "assign", "conditional", "seq" ]) ||
|
|
lvalue[0] == "binary" && PRECEDENCE[operator] > PRECEDENCE[lvalue[1]] ||
|
|
lvalue[0] == "function" && needs_parens(this)) {
|
|
left = "(" + left + ")";
|
|
}
|
|
if (member(rvalue[0], [ "assign", "conditional", "seq" ]) ||
|
|
rvalue[0] == "binary" && PRECEDENCE[operator] >= PRECEDENCE[rvalue[1]] &&
|
|
!(rvalue[1] == operator && member(operator, [ "&&", "||", "*" ]))) {
|
|
right = "(" + right + ")";
|
|
}
|
|
else if (!beautify && options.inline_script && (operator == "<" || operator == "<<")
|
|
&& rvalue[0] == "regexp" && /^script/i.test(rvalue[1])) {
|
|
right = " " + right;
|
|
}
|
|
return add_spaces([ left, operator, right ]);
|
|
},
|
|
"unary-prefix": function(operator, expr) {
|
|
var val = make(expr);
|
|
if (!(expr[0] == "num" || (expr[0] == "unary-prefix" && !HOP(OPERATORS, operator + expr[1])) || !needs_parens(expr)))
|
|
val = "(" + val + ")";
|
|
return operator + (jsp.is_alphanumeric_char(operator.charAt(0)) ? " " : "") + val;
|
|
},
|
|
"unary-postfix": function(operator, expr) {
|
|
var val = make(expr);
|
|
if (!(expr[0] == "num" || (expr[0] == "unary-postfix" && !HOP(OPERATORS, operator + expr[1])) || !needs_parens(expr)))
|
|
val = "(" + val + ")";
|
|
return val + operator;
|
|
},
|
|
"sub": function(expr, subscript) {
|
|
var hash = make(expr);
|
|
if (needs_parens(expr))
|
|
hash = "(" + hash + ")";
|
|
return hash + "[" + make(subscript) + "]";
|
|
},
|
|
"object": function(props) {
|
|
var obj_needs_parens = needs_parens(this);
|
|
if (props.length == 0)
|
|
return obj_needs_parens ? "({})" : "{}";
|
|
var out = "{" + newline + with_indent(function(){
|
|
return MAP(props, function(p){
|
|
if (p.length == 3) {
|
|
// getter/setter. The name is in p[0], the arg.list in p[1][2], the
|
|
// body in p[1][3] and type ("get" / "set") in p[2].
|
|
return indent(make_function(p[0], p[1][2], p[1][3], p[2]));
|
|
}
|
|
var key = p[0], val = parenthesize(p[1], "seq");
|
|
if (options.quote_keys) {
|
|
key = encode_string(key);
|
|
} else if ((typeof key == "number" || !beautify && +key + "" == key)
|
|
&& parseFloat(key) >= 0) {
|
|
key = make_num(+key);
|
|
} else if (!is_identifier(key)) {
|
|
key = encode_string(key);
|
|
}
|
|
return indent(add_spaces(beautify && options.space_colon
|
|
? [ key, ":", val ]
|
|
: [ key + ":", val ]));
|
|
}).join("," + newline);
|
|
}) + newline + indent("}");
|
|
return obj_needs_parens ? "(" + out + ")" : out;
|
|
},
|
|
"regexp": function(rx, mods) {
|
|
return "/" + rx + "/" + mods;
|
|
},
|
|
"array": function(elements) {
|
|
if (elements.length == 0) return "[]";
|
|
return add_spaces([ "[", add_commas(MAP(elements, function(el, i){
|
|
if (!beautify && el[0] == "atom" && el[1] == "undefined") return i === elements.length - 1 ? "," : "";
|
|
return parenthesize(el, "seq");
|
|
})), "]" ]);
|
|
},
|
|
"stat": function(stmt) {
|
|
return make(stmt).replace(/;*\s*$/, ";");
|
|
},
|
|
"seq": function() {
|
|
return add_commas(MAP(slice(arguments), make));
|
|
},
|
|
"label": function(name, block) {
|
|
return add_spaces([ make_name(name), ":", make(block) ]);
|
|
},
|
|
"with": function(expr, block) {
|
|
return add_spaces([ "with", "(" + make(expr) + ")", make(block) ]);
|
|
},
|
|
"atom": function(name) {
|
|
return make_name(name);
|
|
}
|
|
}, function(){ return make(ast) });
|
|
|
|
// The squeezer replaces "block"-s that contain only a single
|
|
// statement with the statement itself; technically, the AST
|
|
// is correct, but this can create problems when we output an
|
|
// IF having an ELSE clause where the THEN clause ends in an
|
|
// IF *without* an ELSE block (then the outer ELSE would refer
|
|
// to the inner IF). This function checks for this case and
|
|
// adds the block brackets if needed.
|
|
function make_then(th) {
|
|
if (th == null) return ";";
|
|
if (th[0] == "do") {
|
|
// https://github.com/mishoo/UglifyJS/issues/#issue/57
|
|
// IE croaks with "syntax error" on code like this:
|
|
// if (foo) do ... while(cond); else ...
|
|
// we need block brackets around do/while
|
|
return make_block([ th ]);
|
|
}
|
|
var b = th;
|
|
while (true) {
|
|
var type = b[0];
|
|
if (type == "if") {
|
|
if (!b[3])
|
|
// no else, we must add the block
|
|
return make([ "block", [ th ]]);
|
|
b = b[3];
|
|
}
|
|
else if (type == "while" || type == "do") b = b[2];
|
|
else if (type == "for" || type == "for-in") b = b[4];
|
|
else break;
|
|
}
|
|
return make(th);
|
|
};
|
|
|
|
function make_function(name, args, body, keyword) {
|
|
var out = keyword || "function";
|
|
if (name) {
|
|
out += " " + make_name(name);
|
|
}
|
|
out += "(" + add_commas(MAP(args, make_name)) + ")";
|
|
out = add_spaces([ out, make_block(body) ]);
|
|
return needs_parens(this) ? "(" + out + ")" : out;
|
|
};
|
|
|
|
function must_has_semicolon(node) {
|
|
switch (node[0]) {
|
|
case "with":
|
|
case "while":
|
|
return empty(node[2]); // `with' or `while' with empty body?
|
|
case "for":
|
|
case "for-in":
|
|
return empty(node[4]); // `for' with empty body?
|
|
case "if":
|
|
if (empty(node[2]) && !node[3]) return true; // `if' with empty `then' and no `else'
|
|
if (node[3]) {
|
|
if (empty(node[3])) return true; // `else' present but empty
|
|
return must_has_semicolon(node[3]); // dive into the `else' branch
|
|
}
|
|
return must_has_semicolon(node[2]); // dive into the `then' branch
|
|
}
|
|
};
|
|
|
|
function make_block_statements(statements, noindent) {
|
|
for (var a = [], last = statements.length - 1, i = 0; i <= last; ++i) {
|
|
var stat = statements[i];
|
|
var code = make(stat);
|
|
if (code != ";") {
|
|
if (!beautify && i == last && !must_has_semicolon(stat)) {
|
|
code = code.replace(/;+\s*$/, "");
|
|
}
|
|
a.push(code);
|
|
}
|
|
}
|
|
return noindent ? a : MAP(a, indent);
|
|
};
|
|
|
|
function make_switch_block(body) {
|
|
var n = body.length;
|
|
if (n == 0) return "{}";
|
|
return "{" + newline + MAP(body, function(branch, i){
|
|
var has_body = branch[1].length > 0, code = with_indent(function(){
|
|
return indent(branch[0]
|
|
? add_spaces([ "case", make(branch[0]) + ":" ])
|
|
: "default:");
|
|
}, 0.5) + (has_body ? newline + with_indent(function(){
|
|
return make_block_statements(branch[1]).join(newline);
|
|
}) : "");
|
|
if (!beautify && has_body && i < n - 1)
|
|
code += ";";
|
|
return code;
|
|
}).join(newline) + newline + indent("}");
|
|
};
|
|
|
|
function make_block(statements) {
|
|
if (!statements) return ";";
|
|
if (statements.length == 0) return "{}";
|
|
return "{" + newline + with_indent(function(){
|
|
return make_block_statements(statements).join(newline);
|
|
}) + newline + indent("}");
|
|
};
|
|
|
|
function make_1vardef(def) {
|
|
var name = def[0], val = def[1];
|
|
if (val != null)
|
|
name = add_spaces([ make_name(name), "=", parenthesize(val, "seq") ]);
|
|
return name;
|
|
};
|
|
|
|
};
|
|
|
|
function split_lines(code, max_line_length) {
|
|
var splits = [ 0 ];
|
|
jsp.parse(function(){
|
|
var next_token = jsp.tokenizer(code);
|
|
var last_split = 0;
|
|
var prev_token;
|
|
function current_length(tok) {
|
|
return tok.pos - last_split;
|
|
};
|
|
function split_here(tok) {
|
|
last_split = tok.pos;
|
|
splits.push(last_split);
|
|
};
|
|
function custom(){
|
|
var tok = next_token.apply(this, arguments);
|
|
out: {
|
|
if (prev_token) {
|
|
if (prev_token.type == "keyword") break out;
|
|
}
|
|
if (current_length(tok) > max_line_length) {
|
|
switch (tok.type) {
|
|
case "keyword":
|
|
case "atom":
|
|
case "name":
|
|
case "punc":
|
|
split_here(tok);
|
|
break out;
|
|
}
|
|
}
|
|
}
|
|
prev_token = tok;
|
|
return tok;
|
|
};
|
|
custom.context = function() {
|
|
return next_token.context.apply(this, arguments);
|
|
};
|
|
return custom;
|
|
}());
|
|
return splits.map(function(pos, i){
|
|
return code.substring(pos, splits[i + 1] || code.length);
|
|
}).join("\n");
|
|
};
|
|
|
|
/* -----[ Utilities ]----- */
|
|
|
|
function repeat_string(str, i) {
|
|
if (i <= 0) return "";
|
|
if (i == 1) return str;
|
|
var d = repeat_string(str, i >> 1);
|
|
d += d;
|
|
if (i & 1) d += str;
|
|
return d;
|
|
};
|
|
|
|
function defaults(args, defs) {
|
|
var ret = {};
|
|
if (args === true)
|
|
args = {};
|
|
for (var i in defs) if (HOP(defs, i)) {
|
|
ret[i] = (args && HOP(args, i)) ? args[i] : defs[i];
|
|
}
|
|
return ret;
|
|
};
|
|
|
|
function is_identifier(name) {
|
|
return /^[a-z_$][a-z0-9_$]*$/i.test(name)
|
|
&& name != "this"
|
|
&& !HOP(jsp.KEYWORDS_ATOM, name)
|
|
&& !HOP(jsp.RESERVED_WORDS, name)
|
|
&& !HOP(jsp.KEYWORDS, name);
|
|
};
|
|
|
|
function HOP(obj, prop) {
|
|
return Object.prototype.hasOwnProperty.call(obj, prop);
|
|
};
|
|
|
|
// some utilities
|
|
|
|
var MAP;
|
|
|
|
(function(){
|
|
MAP = function(a, f, o) {
|
|
var ret = [], top = [], i;
|
|
function doit() {
|
|
var val = f.call(o, a[i], i);
|
|
if (val instanceof AtTop) {
|
|
val = val.v;
|
|
if (val instanceof Splice) {
|
|
top.push.apply(top, val.v);
|
|
} else {
|
|
top.push(val);
|
|
}
|
|
}
|
|
else if (val != skip) {
|
|
if (val instanceof Splice) {
|
|
ret.push.apply(ret, val.v);
|
|
} else {
|
|
ret.push(val);
|
|
}
|
|
}
|
|
};
|
|
if (a instanceof Array) for (i = 0; i < a.length; ++i) doit();
|
|
else for (i in a) if (HOP(a, i)) doit();
|
|
return top.concat(ret);
|
|
};
|
|
MAP.at_top = function(val) { return new AtTop(val) };
|
|
MAP.splice = function(val) { return new Splice(val) };
|
|
var skip = MAP.skip = {};
|
|
function AtTop(val) { this.v = val };
|
|
function Splice(val) { this.v = val };
|
|
})();
|
|
|
|
/* -----[ Exports ]----- */
|
|
|
|
exports.ast_walker = ast_walker;
|
|
exports.ast_mangle = ast_mangle;
|
|
exports.ast_squeeze = ast_squeeze;
|
|
exports.ast_lift_variables = ast_lift_variables;
|
|
exports.gen_code = gen_code;
|
|
exports.ast_add_scope = ast_add_scope;
|
|
exports.set_logger = function(logger) { warn = logger };
|
|
exports.make_string = make_string;
|
|
exports.split_lines = split_lines;
|
|
exports.MAP = MAP;
|
|
|
|
// keep this last!
|
|
exports.ast_squeeze_more = require("./squeeze-more").ast_squeeze_more;
|