CLI source map options
Terser can generate a source map file, which is highly useful for
debugging your compressed JavaScript. To get a source map, pass
--source-map --output output.js
(source map will be written out to
output.js.map
).
Additional options:
--source-map "filename='<NAME>'"
to specify the name of the source map.--source-map "root='<URL>'"
to pass the URL where the original files can be found.--source-map "url='<URL>'"
to specify the URL where the source map can be found. Otherwise Terser assumes HTTPX-SourceMap
is being used and will omit the//# sourceMappingURL=
directive.
For example:
terser js/file1.js js/file2.js \
-o foo.min.js -c -m \
--source-map "root='http://foo.com/src',url='foo.min.js.map'"
The above will compress and mangle file1.js
and file2.js
, will drop the
output in foo.min.js
and the source map in foo.min.js.map
. The source
mapping will refer to http://foo.com/src/js/file1.js
and
http://foo.com/src/js/file2.js
(in fact it will list http://foo.com/src
as the source map root, and the original files as js/file1.js
and
js/file2.js
).
Composed source map
When you’re compressing JS code that was output by a compiler such as CoffeeScript, mapping to the JS code won’t be too helpful. Instead, you’d like to map back to the original code (i.e. CoffeeScript). Terser has an option to take an input source map. Assuming you have a mapping from CoffeeScript → compiled JS, Terser can generate a map from CoffeeScript → compressed JS by mapping every token in the compiled JS to its original location.
To use this feature pass --source-map "content='/path/to/input/source.map'"
or --source-map "content=inline"
if the source map is included inline with
the sources.
CLI compress options
You need to pass --compress
(-c
) to enable the compressor. Optionally
you can pass a comma-separated list of compress options.
Options are in the form foo=bar
, or just foo
(the latter implies
a boolean option that you want to set true
; it’s effectively a
shortcut for foo=true
).
Example:
terser file.js -c toplevel,sequences=false
CLI mangling property names (--mangle-props
)
Note: THIS WILL BREAK YOUR CODE. A good rule of thumb is not to use this unless you know exactly what you’re doing and how this works and read this section until the end.
Mangling property names is a separate step, different from variable name mangling. Pass
--mangle-props
to enable it. The least dangerous
way to use this is to use the regex
option like so:
terser example.js -c -m --mangle-props regex=/_$/
This will mangle all properties that end with an underscore. So you can use it to mangle internal methods.
By default, it will mangle all properties in the input code with the exception of built in DOM properties and properties in core JavaScript classes, which is what will break your code if you don’t:
- Control all the code you’re mangling
- Avoid using a module bundler, as they usually will call Terser on each file individually, making it impossible to pass mangled objects between modules.
- Avoid calling functions like
defineProperty
orhasOwnProperty
, because they refer to object properties using strings and will break your code if you don’t know what you are doing.
An example:
// example.js
var x = {
baz_: 0,
foo_: 1,
calc: function() {
return this.foo_ + this.baz_;
}
};
x.bar_ = 2;
x["baz_"] = 3;
console.log(x.calc());
Mangle all properties (except for JavaScript builtins
) (very unsafe):
$ terser example.js -c passes=2 -m --mangle-props
var x={o:3,t:1,i:function(){return this.t+this.o},s:2};console.log(x.i());
Mangle all properties except for reserved
properties (still very unsafe):
$ terser example.js -c passes=2 -m --mangle-props reserved=[foo_,bar_]
var x={o:3,foo_:1,t:function(){return this.foo_+this.o},bar_:2};console.log(x.t());
Mangle all properties matching a regex
(not as unsafe but still unsafe):
$ terser example.js -c passes=2 -m --mangle-props regex=/_$/
var x={o:3,t:1,calc:function(){return this.t+this.o},i:2};console.log(x.calc());
Combining mangle properties options:
$ terser example.js -c passes=2 -m --mangle-props regex=/_$/,reserved=[bar_]
var x={o:3,t:1,calc:function(){return this.t+this.o},bar_:2};console.log(x.calc());
In order for this to be of any use, we avoid mangling standard JS names and DOM
API properties by default (--mangle-props builtins
to override).
A regular expression can be used to define which property names should be
mangled. For example, --mangle-props regex=/^_/
will only mangle property
names that start with an underscore.
When you compress multiple files using this option, in order for them to
work together in the end we need to ensure somehow that one property gets
mangled to the same name in all of them. For this, pass --name-cache filename.json
and Terser will maintain these mappings in a file which can then be reused.
It should be initially empty. Example:
$ rm -f /tmp/cache.json # start fresh
$ terser file1.js file2.js --mangle-props --name-cache /tmp/cache.json -o part1.js
$ terser file3.js file4.js --mangle-props --name-cache /tmp/cache.json -o part2.js
Now, part1.js
and part2.js
will be consistent with each other in terms
of mangled property names.
Using the name cache is not necessary if you compress all your files in a single call to Terser.
Mangling unquoted names (--mangle-props keep_quoted
)
Using quoted property name (o["foo"]
) reserves the property name (foo
)
so that it is not mangled throughout the entire script even when used in an
unquoted style (o.foo
). Example:
// stuff.js
var o = {
"foo": 1,
bar: 3
};
o.foo += o.bar;
console.log(o.foo);
$ terser stuff.js --mangle-props keep_quoted -c -m
var o={foo:1,o:3};o.foo+=o.o,console.log(o.foo);
Debugging property name mangling
You can also pass --mangle-props debug
in order to mangle property names
without completely obscuring them. For example the property o.foo
would mangle to o._$foo$_
with this option. This allows property mangling
of a large codebase while still being able to debug the code and identify
where mangling is breaking things.
$ terser stuff.js --mangle-props debug -c -m
var o={_$foo$_:1,_$bar$_:3};o._$foo$_+=o._$bar$_,console.log(o._$foo$_);
You can also pass a custom suffix using --mangle-props debug=XYZ
. This would then
mangle o.foo
to o._$foo$XYZ_
. You can change this each time you compile a
script to identify how a property got mangled. One technique is to pass a
random number on every compile to simulate mangling changing with different
inputs (e.g. as you update the input script with new properties), and to help
identify mistakes like writing mangled keys to storage.
Minify options structure
{
parse: {
// parse options
},
compress: {
// compress options
},
mangle: {
// mangle options
properties: {
// mangle property options
}
},
format: {
// format options (can also use `output` for backwards compatibility)
},
sourceMap: {
// source map options
},
ecma: 5, // specify one of: 5, 2015, 2016, etc.
keep_classnames: false,
keep_fnames: false,
ie8: false,
module: false,
nameCache: null, // or specify a name cache object
safari10: false,
toplevel: false,
}
Source map options
To generate a source map:
var result = await minify({"file1.js": "var a = function() {};"}, {
sourceMap: {
filename: "out.js",
url: "out.js.map"
}
});
console.log(result.code); // minified output
console.log(result.map); // source map
Note that the source map is not saved in a file, it’s just returned in
result.map
. The value passed for sourceMap.url
is only used to set
//# sourceMappingURL=out.js.map
in result.code
. The value of
filename
is only used to set file
attribute (see the spec)
in source map file.
You can set option sourceMap.url
to be "inline"
and source map will
be appended to code.
You can also specify sourceRoot property to be included in source map:
var result = await minify({"file1.js": "var a = function() {};"}, {
sourceMap: {
root: "http://example.com/src",
url: "out.js.map"
}
});
If you’re compressing compiled JavaScript and have a source map for it, you
can use sourceMap.content
:
var result = await minify({"compiled.js": "compiled code"}, {
sourceMap: {
content: "content from compiled.js.map",
url: "minified.js.map"
}
});
// same as before, it returns `code` and `map`
If you’re using the X-SourceMap
header instead, you can just omit sourceMap.url
.
If you happen to need the source map as a raw object, set sourceMap.asObject
to true
.
Compress options
defaults
(default:true
) – Passfalse
to disable most default enabledcompress
transforms. Useful when you only want to enable a fewcompress
options while disabling the rest.arrows
(default:true
) – Class and object literal methods are converted will also be converted to arrow expressions if the resultant code is shorter:m(){return x}
becomesm:()=>x
. To do this to regular ES5 functions which don’t usethis
orarguments
, seeunsafe_arrows
.arguments
(default:false
) – replacearguments[index]
with function parameter name whenever possible.booleans
(default:true
) – various optimizations for boolean context, for example!!a ? b : c → a ? b : c
booleans_as_integers
(default:false
) – Turn booleans into 0 and 1, also makes comparisons with booleans use==
and!=
instead of===
and!==
.collapse_vars
(default:true
) – Collapse single-use non-constant variables, side effects permitting.comparisons
(default:true
) – apply certain optimizations to binary nodes, e.g.!(a <= b) → a > b
(only whenunsafe_comps
), attempts to negate binary nodes, e.g.a = !b && !c && !d && !e → a=!(b||c||d||e)
etc.computed_props
(default:true
) – Transforms constant computed properties into regular ones:{["computed"]: 1}
is converted to{computed: 1}
.conditionals
(default:true
) – apply optimizations forif
-s and conditional expressionsdead_code
(default:true
) – remove unreachable codedirectives
(default:true
) – remove redundant or non-standard directivesdrop_console
(default:false
) – Passtrue
to discard calls toconsole.*
functions. If you wish to drop a specific function call such asconsole.info
and/or retain side effects from function arguments after dropping the function call then usepure_funcs
instead.drop_debugger
(default:true
) – removedebugger;
statementsecma
(default:5
) – Pass2015
or greater to enablecompress
options that will transform ES5 code into smaller ES6+ equivalent forms.evaluate
(default:true
) – attempt to evaluate constant expressionsexpression
(default:false
) – Passtrue
to preserve completion values from terminal statements withoutreturn
, e.g. in bookmarklets.global_defs
(default:{}
) – see conditional compilationhoist_funs
(default:false
) – hoist function declarationshoist_props
(default:true
) – hoist properties from constant object and array literals into regular variables subject to a set of constraints. For example:var o={p:1, q:2}; f(o.p, o.q);
is converted tof(1, 2);
. Note:hoist_props
works best withmangle
enabled, thecompress
optionpasses
set to2
or higher, and thecompress
optiontoplevel
enabled.hoist_vars
(default:false
) – hoistvar
declarations (this isfalse
by default because it seems to increase the size of the output in general)if_return
(default:true
) – optimizations for if/return and if/continueinline
(default:true
) – inline calls to function with simple/return
statement:false
– same as0
0
– disabled inlining1
– inline simple functions2
– inline functions with arguments3
– inline functions with arguments and variablestrue
– same as3
join_vars
(default:true
) – join consecutivevar
statementskeep_classnames
(default:false
) – Passtrue
to prevent the compressor from discarding class names. Pass a regular expression to only keep class names matching that regex. See also: thekeep_classnames
mangle option.keep_fargs
(default:true
) – Prevents the compressor from discarding unused function arguments. You need this for code which relies onFunction.length
.keep_fnames
(default:false
) – Passtrue
to prevent the compressor from discarding function names. Pass a regular expression to only keep function names matching that regex. Useful for code relying onFunction.prototype.name
. See also: thekeep_fnames
mangle option.keep_infinity
(default:false
) – Passtrue
to preventInfinity
from being compressed into1/0
, which may cause performance issues on Chrome.loops
(default:true
) – optimizations fordo
,while
andfor
loops when we can statically determine the condition.module
(defaultfalse
) – Passtrue
when compressing an ES6 module. Strict mode is implied and thetoplevel
option as well.negate_iife
(default:true
) – negate “Immediately-Called Function Expressions” where the return value is discarded, to avoid the parens that the code generator would insert.passes
(default:1
) – The maximum number of times to run compress. In some cases more than one pass leads to further compressed code. Keep in mind more passes will take more time.properties
(default:true
) – rewrite property access using the dot notation, for examplefoo["bar"] → foo.bar
pure_funcs
(default:null
) – You can pass an array of names and Terser will assume that those functions do not produce side effects. DANGER: will not check if the name is redefined in scope. An example case here, for instancevar q = Math.floor(a/b)
. If variableq
is not used elsewhere, Terser will drop it, but will still keep theMath.floor(a/b)
, not knowing what it does. You can passpure_funcs: [ 'Math.floor' ]
to let it know that this function won’t produce any side effect, in which case the whole statement would get discarded. The current implementation adds some overhead (compression will be slower).pure_getters
(default:"strict"
) – If you passtrue
for this, Terser will assume that object property access (e.g.foo.bar
orfoo["bar"]
) doesn’t have any side effects. Specify"strict"
to treatfoo.bar
as side-effect-free only whenfoo
is certain to not throw, i.e. notnull
orundefined
.reduce_funcs
(legacy option, safely ignored for backwards compatibility).reduce_vars
(default:true
) – Improve optimization on variables assigned with and used as constant values.sequences
(default:true
) – join consecutive simple statements using the comma operator. May be set to a positive integer to specify the maximum number of consecutive comma sequences that will be generated. If this option is set totrue
then the defaultsequences
limit is200
. Set option tofalse
or0
to disable. The smallestsequences
length is2
. Asequences
value of1
is grandfathered to be equivalent totrue
and as such means200
. On rare occasions the default sequences limit leads to very slow compress times in which case a value of20
or less is recommended.side_effects
(default:true
) – Remove expressions which have no side effects and whose results aren’t used.switches
(default:true
) – de-duplicate and remove unreachableswitch
branchestoplevel
(default:false
) – drop unreferenced functions ("funcs"
) and/or variables ("vars"
) in the top level scope (false
by default,true
to drop both unreferenced functions and variables)top_retain
(default:null
) – prevent specific toplevel functions and variables fromunused
removal (can be array, comma-separated, RegExp or function. Impliestoplevel
)typeofs
(default:true
) – Transformstypeof foo == "undefined"
intofoo === void 0
. Note: recommend to set this value tofalse
for IE10 and earlier versions due to known issues.unsafe
(default:false
) – apply “unsafe” transformations (details).unsafe_arrows
(default:false
) – Convert ES5 style anonymous function expressions to arrow functions if the function body does not referencethis
. Note: it is not always safe to perform this conversion if code relies on the the function having aprototype
, which arrow functions lack. This transform requires that theecma
compress option is set to2015
or greater.unsafe_comps
(default:false
) – Reverse<
and<=
to>
and>=
to allow improved compression. This might be unsafe when an at least one of two operands is an object with computed values due the use of methods likeget
, orvalueOf
. This could cause change in execution order after operands in the comparison are switching. Compression only works if bothcomparisons
andunsafe_comps
are both set to true.unsafe_Function
(default:false
) – compress and mangleFunction(args, code)
when bothargs
andcode
are string literals.unsafe_math
(default:false
) – optimize numerical expressions like2 * x * 3
into6 * x
, which may give imprecise floating point results.unsafe_symbols
(default:false
) – removes keys from native Symbol declarations, e.gSymbol("kDog")
becomesSymbol()
.unsafe_methods
(default: false) – Converts{ m: function(){} }
to{ m(){} }
.ecma
must be set to6
or greater to enable this transform. Ifunsafe_methods
is a RegExp then key/value pairs with keys matching the RegExp will be converted to concise methods. Note: if enabled there is a risk of getting a “<method name>
is not a constructor” TypeError should any code try tonew
the former function.unsafe_proto
(default:false
) – optimize expressions likeArray.prototype.slice.call(a)
into[].slice.call(a)
unsafe_regexp
(default:false
) – enable substitutions of variables withRegExp
values the same way as if they are constants.unsafe_undefined
(default:false
) – substitutevoid 0
if there is a variable namedundefined
in scope (variable name will be mangled, typically reduced to a single character)unused
(default:true
) – drop unreferenced functions and variables (simple direct variable assignments do not count as references unless set to"keep_assign"
)
Mangle options
eval
(defaultfalse
) – Passtrue
to mangle names visible in scopes whereeval
orwith
are used.keep_classnames
(defaultfalse
) – Passtrue
to not mangle class names. Pass a regular expression to only keep class names matching that regex. See also: thekeep_classnames
compress option.keep_fnames
(defaultfalse
) – Passtrue
to not mangle function names. Pass a regular expression to only keep class names matching that regex. Useful for code relying onFunction.prototype.name
. See also: thekeep_fnames
compress option.module
(defaultfalse
) – Passtrue
an ES6 modules, where the toplevel scope is not the global scope. Impliestoplevel
.reserved
(default[]
) – Pass an array of identifiers that should be excluded from mangling. Example:["foo", "bar"]
.toplevel
(defaultfalse
) – Passtrue
to mangle names declared in the top level scope.safari10
(defaultfalse
) – Passtrue
to work around the Safari 10 loop iterator bug “Cannot declare a let variable twice”. See also: thesafari10
format option.
Examples:
// test.js
var globalVar;
function funcName(firstLongName, anotherLongName) {
var myVariable = firstLongName + anotherLongName;
}
var code = fs.readFileSync("test.js", "utf8");
await minify(code).code;
// 'function funcName(a,n){}var globalVar;'
await minify(code, { mangle: { reserved: ['firstLongName'] } }).code;
// 'function funcName(firstLongName,a){}var globalVar;'
await minify(code, { mangle: { toplevel: true } }).code;
// 'function n(n,a){}var a;'
Mangle properties options
builtins
(default:false
) — Usetrue
to allow the mangling of builtin DOM properties. Not recommended to override this setting.debug
(default:false
) — Mangle names with the original name still present. Pass an empty string""
to enable, or a non-empty string to set the debug suffix.keep_quoted
(default:false
) — Only mangle unquoted property names.true
– Quoted property names are automatically reserved and any unquoted property names will not be mangled."strict"
– Advanced, all unquoted property names are mangled unless explicitly reserved.
regex
(default:null
) — Pass a RegExp literal or pattern string to only mangle property matching the regular expression.reserved
(default:[]
) — Do not mangle property names listed in thereserved
array.undeclared
(default:false
) - Mangle those names when they are accessed as properties of known top level variables but their declarations are never found in input code. May be useful when only minifying parts of a project. See #397 for more details.
Miscellaneous
The unsafe
compress
option
It enables some transformations that might break code logic in certain contrived cases, but should be fine for most code. It assumes that standard built-in ECMAScript functions and classes have not been altered or replaced. You might want to try it on your own code; it should reduce the minified size. Some examples of the optimizations made when this option is enabled:
new Array(1, 2, 3)
orArray(1, 2, 3)
→[ 1, 2, 3 ]
new Object()
→{}
String(exp)
orexp.toString()
→"" + exp
new Object/RegExp/Function/Error/Array (...)
→ we discard thenew
"foo bar".substr(4)
→"bar"
Conditional compilation API
You can also use conditional compilation via the programmatic API. With the difference that the
property name is global_defs
and is a compressor property:
var result = await minify(fs.readFileSync("input.js", "utf8"), {
compress: {
dead_code: true,
global_defs: {
DEBUG: false
}
}
});
To replace an identifier with an arbitrary non-constant expression it is
necessary to prefix the global_defs
key with "@"
to instruct Terser
to parse the value as an expression:
await minify("alert('hello');", {
compress: {
global_defs: {
"@alert": "console.log"
}
}
}).code;
// returns: 'console.log("hello");'
Otherwise it would be replaced as string literal:
await minify("alert('hello');", {
compress: {
global_defs: {
"alert": "console.log"
}
}
}).code;
// returns: '"console.log"("hello");'
Annotations
Annotations in Terser are a way to tell it to treat a certain function call differently. The following annotations are available:
/*@__INLINE__*/
- forces a function to be inlined somewhere./*@__NOINLINE__*/
- Makes sure the called function is not inlined into the call site./*@__PURE__*/
- Marks a function call as pure. That means, it can safely be dropped.
You can use either a @
sign at the start, or a #
.
Here are some examples on how to use them:
/*@__INLINE__*/
function_always_inlined_here()
/*#__NOINLINE__*/
function_cant_be_inlined_into_here()
const x = /*#__PURE__*/i_am_dropped_if_x_is_not_used()
ESTree / SpiderMonkey AST
Terser has its own abstract syntax tree format; for practical reasons we can’t easily change to using the SpiderMonkey AST internally. However, Terser now has a converter which can import a SpiderMonkey AST.
For example Acorn is a super-fast parser that produces a SpiderMonkey AST. It has a small CLI utility that parses one file and dumps the AST in JSON on the standard output. To use Terser to mangle and compress that:
acorn file.js | terser -p spidermonkey -m -c
The -p spidermonkey
option tells Terser that all input files are not
JavaScript, but JS code described in SpiderMonkey AST in JSON. Therefore we
don’t use our own parser in this case, but just transform that AST into our
internal AST.
Use Acorn for parsing
More for fun, I added the -p acorn
option which will use Acorn to do all
the parsing. If you pass this option, Terser will require("acorn")
.
Acorn is really fast (e.g. 250ms instead of 380ms on some 650K code), but converting the SpiderMonkey tree that Acorn produces takes another 150ms so in total it’s a bit more than just using Terser’s own parser.
Terser Fast Minify Mode
It’s not well known, but whitespace removal and symbol mangling accounts
for 95% of the size reduction in minified code for most JavaScript - not
elaborate code transforms. One can simply disable compress
to speed up
Terser builds by 3 to 4 times.
d3.js | size | gzip size | time (s) |
---|---|---|---|
original | 451,131 | 108,733 | - |
terser@3.7.5 mangle=false, compress=false | 316,600 | 85,245 | 0.82 |
terser@3.7.5 mangle=true, compress=false | 220,216 | 72,730 | 1.45 |
terser@3.7.5 mangle=true, compress=true | 212,046 | 70,954 | 5.87 |
babili@0.1.4 | 210,713 | 72,140 | 12.64 |
babel-minify@0.4.3 | 210,321 | 72,242 | 48.67 |
babel-minify@0.5.0-alpha.01eac1c3 | 210,421 | 72,238 | 14.17 |
To enable fast minify mode from the CLI use:
terser file.js -m
To enable fast minify mode with the API use:
await minify(code, { compress: false, mangle: true });
Source maps and debugging
Various compress
transforms that simplify, rearrange, inline and remove code
are known to have an adverse effect on debugging with source maps. This is
expected as code is optimized and mappings are often simply not possible as
some code no longer exists. For highest fidelity in source map debugging
disable the compress
option and just use mangle
.
Compiler assumptions
To allow for better optimizations, the compiler makes various assumptions:
.toString()
and.valueOf()
don’t have side effects, and for built-in objects they have not been overridden.undefined
,NaN
andInfinity
have not been externally redefined.arguments.callee
,arguments.caller
andFunction.prototype.caller
are not used.- The code doesn’t expect the contents of
Function.prototype.toString()
orError.prototype.stack
to be anything in particular. - Getting and setting properties on a plain object does not cause other side effects
(using
.watch()
orProxy
). - Object properties can be added, removed and modified (not prevented with
Object.defineProperty()
,Object.defineProperties()
,Object.freeze()
,Object.preventExtensions()
orObject.seal()
). document.all
is not== null
- Assigning properties to a class doesn’t have side effects and does not throw.
Build Tools and Adaptors using Terser
https://www.npmjs.com/browse/depended/terser
Replacing uglify-es
with terser
in a project using yarn
A number of JS bundlers and uglify wrappers are still using buggy versions
of uglify-es
and have not yet upgraded to terser
. If you are using yarn
you can add the following alias to your project’s package.json
file:
"resolutions": {
"uglify-es": "npm:terser"
}
to use terser
instead of uglify-es
in all deeply nested dependencies
without changing any code.
Note: for this change to take effect you must run the following commands
to remove the existing yarn
lock file and reinstall all packages:
$ rm -rf node_modules yarn.lock
$ yarn
Contributors
Financial Contributors
Become a financial contributor and help us sustain our community. [Contribute]