* Compile classes to ES2015 classes
Rather than compiling classes to named functions with prototype and
class assignments, they are now compiled to ES2015 class declarations.
Backwards compatibility has been maintained by compiling ES2015-
incompatible properties as prototype or class assignments. `super`
continues to be compiled as before.
Where possible, classes will be compiled "bare", without an enclosing
IIFE. This is possible when the class contains only ES2015 compatible
expressions (methods and static methods), and has no parent (this last
constraint is a result of the legacy `super` compilation, and could be
removed once ES2015 `super` is being used). Classes are still assigned
to variables to maintain compatibility for assigned class expressions.
There are a few changes to existing functionality that could break
backwards compatibility:
- Derived constructors that deliberately don't call `super` are no
longer possible. ES2015 derived classes can't use `this` unless the
parent constructor has been called, so it's now called implicitly when
not present.
- As a consequence of the above, derived constructors with @ parameters
or bound methods and explicit `super` calls are not allowed. The
implicit `super` must be used in these cases.
* Add tests to verify class interoperability with ES
* Refactor class nodes to separate executable body logic
Logic has been redistributed amongst the class nodes so that:
- `Class` contains the logic necessary to compile an ES class
declaration.
- `ExecutableClassBody` contains the logic necessary to compile CS'
class extensions that require an executable class body.
`Class` still necessarily contains logic to determine whether an
expression is valid in an ES class initializer or not. If any invalid
expressions are found then `Class` will wrap itself in an
`ExecutableClassBody` when compiling.
* Rename `Code#static` to `Code#isStatic`
This naming is more consistent with other `Code` flags.
* Output anonymous classes when possible
Anonymous classes can be output when:
- The class has no parent. The current super compilation needs a class
variable to reference. This condition will go away when ES2015 super
is in use.
- The class contains no bound static methods. Bound static methods have
their context set to the class name.
* Throw errors at compile time for async or generator constructors
* Improve handling of anonymous classes
Anonymous classes are now always anonymous. If a name is required (e.g.
for bound static methods or derived classes) then the class is compiled
in an `ExecutableClassBody` which will give the anonymous class a stable
reference.
* Add a `replaceInContext` method to `Node`
`replaceInContext` will traverse children looking for a node for which
`match` returns true. Once found, the matching node will be replaced by
the result of calling `replacement`.
* Separate `this` assignments from function parameters
This change has been made to simplify two future changes:
1. Outputting `@`-param assignments after a `super` call.
In this case it is necessary that non-`@` parameters are available
before `super` is called, so destructuring has to happen before
`this` assignment.
2. Compiling destructured assignment to ES6
In this case also destructuring has to happen before `this`,
as destructuring can happen in the arguments list, but `this`
assignment can not.
A bonus side-effect is that default values for `@` params are now output
as ES6 default parameters, e.g.
(@a = 1) ->
becomes
function a (a = 1) {
this.a = a;
}
* Change `super` handling in class constructors
Inside an ES derived constructor (a constructor for a class that extends
another class), it is impossible to access `this` until `super` has been
called. This conflicts with CoffeeScript's `@`-param and bound method
features, which compile to `this` references at the top of a function
body. For example:
class B extends A
constructor: (@param) -> super
method: =>
This would compile to something like:
class B extends A {
constructor (param) {
this.param = param;
this.method = bind(this.method, this);
super(...arguments);
}
}
This would break in an ES-compliant runtime as there are `this`
references before the call to `super`. Before this commit we were
dealing with this by injecting an implicit `super` call into derived
constructors that do not already have an explicit `super` call.
Furthermore, we would disallow explicit `super` calls in derived
constructors that used bound methods or `@`-params, meaning the above
example would need to be rewritten as:
class B extends A
constructor: (@param) ->
method: =>
This would result in a call to `super(...arguments)` being generated as
the first expression in `B#constructor`.
Whilst this approach seems to work pretty well, and is arguably more
convenient than having to manually call `super` when you don't
particularly care about the arguments, it does introduce some 'magic'
and separation from ES, and would likely be a pain point in a project
that made use of significant constructor overriding.
This commit introduces a mechanism through which `super` in constructors
is 'expanded' to include any generated `this` assignments, whilst
retaining the same semantics of a super call. The first example above
now compiles to something like:
class B extends A {
constructor (param) {
var ref
ref = super(...arguments), this.param = param, this.method = bind(this.method, this), ref;
}
}
* Improve `super` handling in constructors
Rather than functions expanding their `super` calls, the `SuperCall`
node can now be given a list of `thisAssignments` to apply when it is
compiled.
This allows us to use the normal compiler machinery to determine whether
the `super` result needs to be cached, whether it appears inline or not,
etc.
* Fix anonymous classes at the top level
Anonymous classes in ES are only valid within expressions. If an
anonymous class is at the top level it will now be wrapped in
parenthses to force it into an expression.
* Re-add Parens wrapper around executable class bodies
This was lost in the refactoring, but it necessary to ensure
`new class ...` works as expected when there's an executable body.
* Throw compiler errors for badly configured derived constructors
Rather than letting them become runtime errors, the following checks are
now performed when compiling a derived constructor:
- The constructor **must** include a call to `super`.
- The constructor **must not** reference `this` in the function body
before `super` has been called.
* Add some tests exercising new class behaviour
- async methods in classes
- `this` access after `super` in extended classes
- constructor super in arrow functions
- constructor functions can't be async
- constructor functions can't be generators
- derived constructors must call super
- derived constructors can't reference `this` before calling super
- generator methods in classes
- 'new' target
* Improve constructor `super` errors
Add a check for `super` in non-extended class constructors, and
explicitly mention derived constructors in the "can't reference this
before super" error.
* Fix compilation of multiple `super` paths in derived constructors
`super` can only be called once, but it can be called conditionally from
multiple locations. The chosen fix is to add the `this` assignments to
every super call.
* Additional class tests, added as a separate file to simplify testing and merging.
Some methods are commented out because they currently throw and I'm not sure how
to test for compilation errors like those.
There is also one test which I deliberately left without passing, `super` in an external prototype override.
This test should 'pass' but is really a variation on the failing `super only allowed in an instance method`
tests above it.
* Changes to the tests. Found bug in super in prototype method. fixed.
* Added failing test back in, dealing with bound functions in external prototype overrides.
* Located a bug in the compiler relating to assertions and escaped ES6 classes.
* Move tests from classes-additional.coffee into classes.coffee; comment out console.log
* Cleaned up tests and made changes based on feedback. Test at the end still has issues, but it's commented out for now.
* Make HoistTarget.expand recursive
It's possible that a hoisted node may itself contain hoisted nodes (e.g.
a class method inside a class method). For this to work the hoisted
fragments need to be expanded recursively.
* Uncomment final test in classes.coffee
The test case now compiles, however another issue is affecting the test
due to the error for `this` before `super` triggering based on source
order rather than execution order. These have been commented out for
now.
* Fixed last test TODOs in test/classes.coffee
Turns out an expression like `this.foo = super()` won't run in JS as it
attempts to lookup `this` before evaluating `super` (i.e. throws "this
is not defined").
* Added more tests for compatability checks, statics, prototypes and ES6 expectations. Cleaned test "nested classes with super".
* Changes to reflect feedback and to comment out issues that will be addressed seperately.
* Clean up test/classes.coffee
- Trim trailing whitespace.
- Rephrase a condition to be more idiomatic.
* Remove check for `super` in derived constructors
In order to be usable at runtime, an extended ES class must call `super`
OR return an alternative object. This check prevented the latter case,
and checking for an alternative return can't be completed statically
without control flow analysis.
* Disallow 'super' in constructor parameter defaults
There are many edge cases when combining 'super' in parameter defaults
with @-parameters and bound functions (and potentially property
initializers in the future).
Rather than attempting to resolve these edge cases, 'super' is now
explicitly disallowed in constructor parameter defaults.
* Disallow @-params in derived constructors without 'super'
@-parameters can't be assigned unless 'super' is called.
* Output interpolated strings as template literals.
* Update comments to use Markdown, so the annotated source reads well; remove TODOs per @lydell’s explanations
Rather than compiling splats to arrays built using `Array#concat`, splats
are now compiled directly to ES2015 splats, e.g.
f foo, arguments..., bar
[ foo, arguments..., bar ]
Which used to be compiled to:
f.apply(null, [foo].concat(slice.call(arguments), [bar]));
[foo].concat(slice.call(arguments), [bar]);
Is now compiled to:
f(foo, ...arguments, bar);
[ foo, ...arguments, bar ];
* Eliminate wrapper around “bound” (arrow) functions; output `=>` for such functions
* Remove irrelevant (and breaking) tests
* Minor cleanup
* When a function parameter is a splat (i.e., it uses the ES2015 rest parameter syntax) output that parameter as ES2015
* Rearrange function parameters when one of the parameters is a splat and isn’t the last parameter (very WIP)
* Handle params like `@param`, adding assignment expressions for them when they appear; ensure splat parameter is last
* Add parameter names (not a text like `'\nValue IdentifierLiteral: a'`) to the scope, so that parameters can’t be deleted; move body-related lines together; more explanation of what’s going on
* For parameters with a default value, correctly add the parameter name to the function scope
* Handle expansions in function parameters: when an expansion is found, set the parameters to only be the original parameters left of the expansion, then an `...args` parameter; and in the function body define variables for the parameters to the right of the expansion, including setting default values
* Handle splat parameters the same way we handle expansions: if a splat parameter is found, it becomes the last parameter in the function definition, and all following parameters get declared in the function body. Fix the splat/rest parameter values after the post-splat parameters have been extracted from it. Clean up `Code.compileNode` so that we loop through the parameters only once, and we create all expressions using calls like `new IdentifierLiteral` rather than `@makeCode`.
* Fix parameter name when a parameter is a splat attached to `this` (e.g. `@param...`)
* Rather than assigning post-splat parameters based on index, use slice; passes test “Functions with splats being called with too few arguments”
* Dial back our w00t indentation
* Better parsing of parameter names (WIP)
* Refactor processing of splat/expansion parameters
* Fix assignment of default parameters for parameters that come after a splat
* Better check for whether a param is attached to `this`
* More understandable variable names
* For parameters after a splat or expansion, assign them similar to the 1.x destructuring method of using `arguments`, except only concern ourselves with the post-splat parameters instead of all parameters; and use the splat/expansion parameter name, since `arguments` in ES fat arrow functions refers to the parent function’s `arguments` rather than the fat arrow function’s arguments/parameters
* Don’t add unnamed parameters (like `[]` as a parameter) to the function scope
* Disallow multiple splat/expansion parameters in function definitions; disallow lone expansion parameters
* Fix `this` params not getting assigned if the parameter is after a splat parameter
* Allow names of function parameters attached to `this` to be reserved words
* Always add a statement to the function body defining a variable with its default value, if it has one, if the variable `== null`; this covers the case when ES doesn’t apply the default value when `null` is passed in as a value, but CoffeeScript expects `null` and `undefined` to act interchangeably
* Aftermath of having both `undefined` and `null` trigger the use of default values for parameters with default values
* More careful parsing of destructured parameters
* Fall back to processing destructured parameters in the function body, to account for `this` or default values within destructured objects
* Clean up comments
* Restore new bare function test, minus the arrow function part of it
* Test that bound/arrow functions aren’t overwriting the `arguments` object, which should refer to the parent scope’s `arguments` (like `this`)
* Follow ES2015 spec for parameter default values: `null` gets assigned as as `null`, not the default value
* Mimic ES default parameters behavior for parameters after a splat or expansion parameter
* Bound functions cannot be generators: remove no-longer-relevant test, add check to throw error if `yield` appears inside a bound (arrow) function
* Error for bound generator functions should underline the `yield`
Any variables generated by CoffeeScript are now made sure to be named to
something not present in the source code being compiled. This way you can no
longer interfere with them, either on purpose or by mistake. (#1500, #1574)
For example, `({a}, _arg) ->` now compiles correctly. (#1574)
As opposed to the somewhat complex implementations discussed in #1500, this
commit takes a very simple approach by saving all used variables names using a
single pass over the token stream. Any generated variables are then made sure
not to exist in that list.
`(@a) -> a` used to be equivalent to `(@a) -> @a`, but now throws a runtime
`ReferenceError` instead (unless `a` exists in an upper scope of course). (#3318)
`(@a) ->` used to compile to `(function(a) { this.a = a; })`. Now it compiles to
`(function(_at_a) { this.a = _at_a; })`. (But you cannot access `_at_a` either,
of course.)
Because of the above, `(@a, a) ->` is now valid; `@a` and `a` are not duplicate
parameters.
Duplicate this-parameters with a reserved word, such as `(@case, @case) ->`,
used to compile but now throws, just like regular duplicate parameters.
- Fix#3394: Unclosed single-quoted strings (both regular ones and heredocs)
used to pass through the lexer, causing a parsing error later, while
double-quoted strings caused an error already in the lexing phase. Now both
single and double-quoted unclosed strings error out in the lexer (which is the
more logical option) with consistent error messages. This also fixes the last
comment by @satyr in #3301.
- Similar to the above, unclosed heregexes also used to pass through the lexer
and not error until in the parsing phase, which resulted in confusing error
messages. This has been fixed, too.
- Fix#3348, by adding passing tests.
- Fix#3529: If a string starts with an interpolation, an empty string is no
longer emitted before the interpolation (unless it is needed to coerce the
interpolation into a string).
- Block comments cannot contain `*/`. Now the error message also shows exactly
where the offending `*/`. This improvement might seem unrelated, but I had to
touch that code anyway to refactor string and regex related code, and the
change was very trivial. Moreover, it's consistent with the next two points.
- Regexes cannot start with `*`. Now the error message also shows exactly where
the offending `*` is. (It might actually not be exatly at the start in
heregexes.) It is a very minor improvement, but it was trivial to add.
- Octal escapes in strings are forbidden in CoffeeScript (just like in
JavaScript strict mode). However, this used to be the case only for regular
strings. Now they are also forbidden in heredocs. Moreover, the errors now
point at the offending octal escape.
- Invalid regex flags are no longer allowed. This includes repeated modifiers
and unknown ones. Moreover, invalid modifiers do not stop a heregex from
being matched, which results in better error messages.
- Fix#3621: `///a#{1}///` compiles to `RegExp("a" + 1)`. So does
`RegExp("a#{1}")`. Still, those two code snippets used to generate different
tokens, which is a bit weird, but more importantly causes problems for
coffeelint (see clutchski/coffeelint#340). This required lots of tests in
test/location.coffee to be updated. Note that some updates to those tests are
unrelated to this point; some have been updated to be more consistent (I
discovered this because the refactored code happened to be seemingly more
correct).
- Regular regex literals used to erraneously allow newlines to be escaped,
causing invalid JavaScript output. This has been fixed.
- Heregexes may now be completely empty (`//////`), instead of erroring out with
a confusing message.
- Fix#2388: Heredocs and heregexes used to be lexed simply, which meant that
you couldn't nest a heredoc within a heredoc (double-quoted, that is) or a
heregex inside a heregex.
- Fix#2321: If you used division inside interpolation and then a slash later in
the string containing that interpolation, the division slash and the latter
slash was erraneously matched as a regex. This has been fixed.
- Indentation inside interpolations in heredocs no longer affect how much
indentation is removed from each line of the heredoc (which is more
intuitive).
- Whitespace is now correctly trimmed from the start and end of strings in a few
edge cases.
- Last but not least, the lexing of interpolated strings now seems to be more
efficient. For a regular double-quoted string, we used to use a custom
function to find the end of it (taking interpolations and interpolations
within interpolations etc. into account). Then we used to re-find the
interpolations and recursively lex their contents. In effect, the same string
was processed twice, or even more in the case of deeper nesting of
interpolations. Now the same string is processed just once.
- Code duplication between regular strings, heredocs, regular regexes and
heregexes has been reduced.
- The above two points should result in more easily read code, too.
