Fix bug where a `throw` was generated without expression
Reenable try/catch/finally and fix them up
Skip serialization errors
Allow function decl in other funcs but not in blocks etc
Rename function to be more appropriate
Fix global functions not getting certain names
Make the canaries more likely to appear as expressions
Add a silly rounding edge case
Add a new canary, `c`, which should only ever be incremented
Refactoring
Fix (another) iife not actually being invoked
When a statement hits recursion max return an expression instead of `;`
When a expression hits recursion max also inc `c`
Generate global code as well as function code
Also fixes some argument juggling related bugs.
No longer reduces the recursion max when generating sub functions.
Generates a function arg.
Add used names to var name pool while in that scope
This is a little wonky, possibly a hack, but since it's synchronous code I think it's alright to do this. The alternative is to slice the varnames array and juggle them through almost all the generator functions and there are various reasons why this patch is a better alternative.
Minify generated code, not beautified code. Prevents beautifier bias.
Prevent unnecessary duplication
Remove serialization protection because I think it got handled elsewhere
Abstract toplevel code generation
Add example line of running test case
Add poor man options parser, and some options
Reindent to 4 spaces
Lower chance of `default` generation
Comment example of testing a case and output improvement
Enable `default` clause appearing at any clause index
Removing some training wheels; dont add parens where we dont absolutely need them
Support `-s1` and `-s2` to force specific statements being generated at that recursion level
Add round number to output when failing. For stats and fun and profit.
Solidify statement depth counting. The argument juggling is real.
Renamed option to something long. -scf was ugly and probably confusing.
Fix missing arguments causing `canThrow` to be truthy, generating crashing code
Generate more binary nested expressions
Add black and white list cli options for statement generation
Allows you to explicitly require or forbid certain statements from/to being made.
```
node test/ufuzz.js --without-stmt switch,try -t 5 -r 5 -V
```
```
node test/ufuzz.js --only-stmt ifelse,expr -t 5 -r 5 -V
```
Similar granularity for expression may be added later.
There can be no comma between names; it just does a split on that arg.
Trim down the binary expression generator
Prevent scoping issues in nodejs by preventing certain names in global space
Oh this list was incomplete?
Allow bin-expr to generate assignments too. More vigilant with storing and reusing vars.
Add more global builtin names
Update wrapper code
Also patch Function valueOf
- move these optimisations out from `Compressor` to `OutputStream`
- fixes behaviour inconsistency when running uglified code from global or module levels due to redefinition
When fixing catch-related issue in #1715, it tries to optimise for duplicate definitions but did not take anonymous functions into account.
Remove such optimisation for now and we can cover this as a more general rule later.
`statement_to_expression()` drops `typeof` even if it operates on undeclared variables.
Since we now have `drop_side_effect_free()`, replace and remove this deprecated functionality.
Unlike normal variables and even function definitions, these cannot be reassigned, even though assignment expressions would "leak" the assigned value as normal.
Process variable definitions with or without assigned values against:
- `arguments`
- named function arguments
- multiple definitions within same scope
Essentially demote variable declarations with no value assignments.
Also fixed invalid use of `AST_VarDef` over `arguments` - should use a member of `AST_SymbolDeclaration` instead.
Conditions including strict mode would make `delete` return `true` or `false`, and are too complex to be evaluated by the compressor.
Suppress assignment folding into said operator.
fixes#1685
@qfox Put value constants in a global constant 74c0fb9
@qfox And the other string based values as well a5033c5
@qfox Be more strict about parameters, allow max to be optional 9c7ce70
@qfox Support a `V` (capital) flag to only log out at intervals 2d822c7
@qfox Fewer magic variables a6a9a7c
@qfox Fix decrement such that a function is created when n=1 7e4b017
@qfox Add more values 64e596e
@qfox Make `b` appear more often d33191a
@qfox Add functions that contain (only..) functions 29a86e3
@qfox Allow the block statement to contain multiple statements 7570484
@qfox Make the interval count a constant d587ad8
@qfox Enable mangling, disable post-processing … 4dc8d35
@qfox Add more simple value that may trigger syntactic errors 8496d58
@qfox Add `else` to some `if` statements a4aed65
@qfox Move iife to expr generator, fix missing recursion arg e453159
@qfox Improve output on error where it wasnt printing the last code properly 4565a1a
@qfox Add switch statement to generator ceafa76
@qfox Add var statement, support optional comma for expr generator b83921b
@qfox Expression generator should use a simple value instead of `0` as recu… … 9d1a5c7
@qfox const -> var to keep things es5... 0143099
@qfox Add more simple values that may trigger edge cases 5e124f1
@qfox Add central name generator, take special care for global functions aeb7682
@qfox Add some `return` and function declaration cases to statement generator 6c9c3cc
@qfox Exclude switches from generator for now 91124b2
Put value constants in a global constant
And the other string based values as well
Be more strict about parameters, allow max to be optional
Support a `V` (capital) flag to only log out at intervals
Fewer magic variables
Fix decrement such that a function is created when n=1
Add more values
Make `b` appear more often
Add functions that contain (only..) functions
Allow the block statement to contain multiple statements
Make the interval count a constant
Enable mangling, disable post-processing
Mangling is kind of the whole point...
Similarly, to beautify the minified code afterwards may supress bugs so it's probably best not to beautify the code prematurely. And there's no point anyways since you won't see it most of the time and only care about the main input anyways.
Add more simple value that may trigger syntactic errors
Add `else` to some `if` statements
Move iife to expr generator, fix missing recursion arg
Improve output on error where it wasnt printing the last code properly
Add switch statement to generator
Add var statement, support optional comma for expr generator
Expression generator should use a simple value instead of `0` as recursion default
const -> var to keep things es5...
Add more simple values that may trigger edge cases
Add central name generator, take special care for global functions
Add some `return` and function declaration cases to statement generator
Exclude switches from generator for now
Enable switch generation because #1667 was merged
Add typeof generator
Add some elision tests
Add a new edge case that returns an object explicitly
Add all binary ops to try and cover more paths
Forgot four binops and added `Math` to var name pool
Harden the incremental pre/postfix tests
Improve switch generator, allow `default` to appear at any clause index
Add try/catch/finally generation
Prevent function statements being generated
Add edge case with decremental op and a group
Disable switch generation until #1679 and #1680 are solved
Only allow `default` clause as last clause for now
Tentatively enable `throw`, `break` and `continue` statements when in valid contexts
`AST_Try` is an `AST_Block`, so besides try block we also need to inspect catch and finally blocks for possible side effects.
Also extend this functionality to handle `AST_If` and `AST_LabeledStatement` while we are at it.
fixes#1673
Turns out the only place in `Compressor` which can generate invalid `AST_For.init` is within `drop_unused()`, so focus the fix-up efforts.
supercedes #1652fixes#1656
- #1634 bars variables with cross-scope references in between to collapse
- but if assigned value is side-effect-free, no states can be modified, so it is safe to move
- mention performance anomaly in Node 7 and drop from CI
- remove unused npm "scripts"
- mark browserify dependency as optional
- stop `test/mozilla-ast.js` from spamming console output in later versions of Node.js
That particular version of Node.js has messed up error messages, so provide a version-specific workaround.
Also fixed an formatting issue which would cause `expect_stdout` to fail if error message contains excerpts of input.
Apply `expect_stdout` to more applicable tests.
The following is wrong:
`a == (b ? a : c)` => `b`
Because:
- `b` may not be boolean
- `a` might have side effects
- `a == a` is not always `true` (think `NaN`)
- `a == c` is not always `false`
`input` and `expect` are parsed as `AST_BlockStatement` which does not support `AST_Directive` by default.
Emulate that by transforming preceding `AST_SimpleStatement`s of `AST_String` into `AST_Directive`.
`reduce_vars` enables substitution of variables but did not clone the value's `AST_Node`.
This confuses `collapse_vars` and result in invalid AST and subsequent crash.
fixes#1609
Liberal use of `Compressor.transform()` and `AST_Node.optimize()` presents an issue for look-up operations like `TreeWalker.in_boolean_context()` and `TreeWalker.parent()`.
This is an incremental fix such that `AST_Node.optimize()` would now contain the correct stack information when called correctly.
Self-referenced function has non-fixed values assigned to its parameters.
Let `unused` & `!keep_fnames` do the scanning, then apply `reduce_vars` only to unnamed functions.
fixes#1595
As patched on `harmony`, `statement()` is the only user of `embed_tokens()` with a missing error branch.
Updated test case and match up with `harmony` to facilitate future merging.
- one-use function call => IIFE should take `eval()` & `arguments` into account
- if unused parameter cannot be eliminated, replace it with `0`
fixes#1583
those were not optimised for `unused` before, which made it necessary for `reduce_vars` to have separate steps for `keep_fnames`
docs update by @kzc
closes#1577
Function expression can be assigned to a variable and be given a name. Ensure function name is the reduced variable before clearing it out.
fixes#1573fixes#1575
- omitting trailing semi-colon in do-while breaks non-browser parser, e.g. uglify-js 1.x
- trailing semi-colon only breaks IE8 if followed by `else` or `while`
- always use braces in do-while body to workaround 2nd case with no size loss in compression
fixes#1568
`AST_Label.references` get `.initialize()` to `[]` every time after `.clone()`
So walk down the tree to pick up the cloned `AST_LoopControl` pieces and put it back together.
`unsafe` turns undefined keyword into a variable of the same name if found, but that interferes with other related optimisations.
Keep track of such transformations to ensure zero information loss in the process.
Bookmarklet for instance implicitedly assumes a "completion value" without using `return`.
The `expression` option now supports such use cases.
Optimisations on IIFEs also enhanced.
fixes#354fixes#543fixes#625fixes#628fixes#640closes#1293
Avoid variable substitution in the following cases:
- use of variable before declaration
- declaration within conditional code blocks
- declaration within loop body
fixes#1518fixes#1525
Modules like webpack and grunt-contrib-uglify still uses `ast.transform(compressor)` before `Compressor.compress(ast)` was introduced.
Workaround this compatibility issue by deactivating `reduce_vars` in such case.
Also fix use case with omitted `options` when calling `Compressor()`.
fixes#1516
A function call or IIFE with an immediately preceding comment
containing `@__PURE__` or `#__PURE__` is deemed to be a
side-effect-free pure function call and can potentially be
dropped.
Depends on `side_effects` option.
`[#@]__PURE__` hint will be removed from comment when pure
call is dropped.
fixes#1261closes#1448
- support arrays, objects & AST_Node
- support `"a.b":1` on both cli & API
- emit warning if variable is modified
- override top-level variables
fixes#1416closes#1198closes#1469
previously test cases with the same name would be skipped except for the last one
`test/run-test.js` will now report duplicated names as errors
closes#1461
- remove extra tree scanning phase for `negate_iife`
- `negate_iife` now only deals with the narrowest form, i.e. IIFE sitting directly under `AST_SimpleStatement`
- `booleans`, `conditionals` etc. will now take care the rest via more accurate accounting
- `a(); void b();` => `a(); b();`
fixes#1288closes#1451
- assign statement does not count towards variable usage by default
- only works with assignments on the same scope level as declaration
- can be disabled with `unused` set to "keep_assign"
- `toplevel` to drop unused top-level variables and/or functions
- `top_retain` to whitelist top-level exceptions
closes#1450
- remove call to evaluate() in is_constant() and let nested optimize() does its job instead
- reject RegExp in is_constant() and remove special case logic under collapse_vars
- operands to conditionals optimisation are now always evaluate()-ed
- throw error in constant_value() instead of returning undefined to catch possible bugs, similar to make_node_from_constant()
- optimise binary boolean operators under `evaluate` instead of `conditionals`
If using `inSourceMap` this fix will ensure the copying of `sourcesContent` is based on potentially normalized `sources` values (https://github.com/mozilla/source-map/blob/master/lib/source-map-consumer.js#L304-L309).
For example `normalize` (https://github.com/mozilla/source-map/blob/master/lib/util.js#L80-L123) will rewrite `./dist/mySource.js` to `dist/mySource.js` in the target `_sources` of the `SourceMapConsumer`. As a result `orig_map.sourceContentFor(source, true);` would return `null` since the orginal `source` was no longer available in the consumer. By using the keys generating from the `SourceMapConsumer.constructor` consistency is ensured.
Using the symbol declaration tracking of UglifyJS doesn't make sense here
anyway, `arguments` always comes from something in the current scope.
fixes#1299
Fixes: #1247
Fix --mangle-props and --name-cache inconsistency.
AST_Dot and AST_Sub properties are now mangled by --mangle-props
without regard to being used in an assignment statement.
Note: if --mangle-props is used then *all* javascript files used must
be uglified with the same mangle options.
Fix the ignore_quoted=true mangle option, also known as
`--mangle-props=unquoted`. If a given property is quoted anywhere
it will not be mangled in any quoted or non-quoted context.
* Escaped newlines should also produce SyntaxError
* Fix multiline comment parsing and add tests
* Adapt makePredicate to handle \u2028 and \u2029
* Move up nlb check in regex so it's checked before any escape handling
* Change error messages to conform ecma standard
* Find_eol not recornizing \u2028 and \u2029 as line terminator
* Remove \u180e as it is removed in unicode 6.3.0 from the category zs
catch identifier is mangled correctly for ES5 standards-compliant JS engines by default.
Unconditionally use the ie8 if/do-while workaround whether or not --screw-ie8 is enabled.
To support non-standard ie8 javascript use: uglifyjs --support-ie8
This should simplify and improve implementation, make it easier to
implement template strings, and keep master a bit more in sync with
harmony.
Previous implementation wasn't broken, though the loop gave me the
impression it could read infinite numbers and annoyed me a bit. It was
also slightly unnecessary because the lookup involved only 3 characters.
The option added in fbbaa42ee5 wasn't
being respected inside object literals, so quoted property names would
still be stripped out with this option.
This is mostly a corner-case, but useful when the output is passed to
something like the Closure compiler, where quoted property names can be
used to prevent mangling.
* Don't interpret strings with escaped content as directive
* Don't interpret strings after empty statement as directive
* Adapt output to prevent strings being represent as directive
* Introduce UGLIFY_DEBUG to allow internal testing like EXPECT_DIRECTIVE
Discarding unused function arguments affects function.length, which can lead
to some hard to debug issues. This optimization is now done only in "unsafe
mode".
Fix#121
In 992b6b9fcc unit test broke (which I missed). This was due to undeclared variables not being side-effects free.
However, since they're really not side-effect free, just declare them in the test cases.
NaN and Infinity were replaced in the output generation, instead of
during compression. This could lead to results where `1/0` was inserted
without parens leading to invalid output.
The nodes are replaced in the compression step now, and the output
generation returns their regular names. This should not be a problem,
since they're already only constructed from the original name.
We can in fact lift sequences, but only if the operation is assignment and
the left-hand side has no side effects nor property access -- that should
guarantee that whatever we place before it cannot affect the sense of the
assignment.
Dropped contrived test case (too hard to support it now), added a more
meaningful one.
By adding the exit code 1 (or any other non-zero exit code) `npm test`
will know the tests didn't perform correctly. This way it's easier to
know if pull requests are good or bad.
ev() should do a single thing — evaluate constant expressions. if that's
not possible, just return the original node. it's not the best place for
partial evaluation there, instead doing it in the compress functions.
1529ab96 started to do this (by considering holes to be separate from
"undefined") but it still converted
[1,,] (length 2, last element hole, trailing comma)
to
[1,] (length 1, trailing comma)
Unfortunately the test suite doesn't really make this clear: the new test here
passes with or without this patch because run-tests.js beautifys the expected
output (in "make_code"), which does the incorrect transformation! If you make
some manual change to arrays.js to make the test fail and see the INPUT and
OUTPUT, then you can see that without this fix, [1,,] -> [1,], and with this fix
it stays [1,,].
1b6bcca7 was a first attempt at this. That commit made Uglify stop replacing
holes with undefined, but instead it started replacing undefined with
holes. This is slightly problematic, because there is a difference between a
hole and an undefined value. More problematically, it changed [1,undefined] to
[1,] which generally doesn't even parse as a hole (just as a trailing comma), so
it didn't even preserve the length of the array!
Instead, parse holes as their own special AST node which prints invisibly.
for(...; x; ...) if (y) break; → for(...; x&&!y; ...);
similarly for `while` and some combinations (i.e. the `break` appears in the
`else` clause, etc.)
- do multiple passes in tighten_body if it was changed
- transform if (foo) return x; return y; ==> return foo?x:y
- don't optimize !0 as true (use best_of after evaluation of constant expr)
With hoist_vars off we now beat UglifyJS v1 on jQuery-1.8.1
