This PR allows legacy decorators to work with strict class fields, which
are now stage 3 and have shipped in a number of browsers. Allowing this will
allow users of the legacy transform (which is currently recommended by the
champions of the decorator proposal) to use the proper class field semantics
for non-decorated fields, which should help prevent breakage later on.
This change is not a breaking change, since users had to explicitly opt into
loose mode in class fields before. This just gives them the option to remove
that opt-in.
* destructuring private fields with array pattern / object pattern
* wip: new test cases
* destrucuring and rest for private properties
* test case for loose private-loose
* add transform-desturcturing for exec
* update test case
* remove getPrototypeOf imports from get and set
* wip: destructure super assignment
* throw "Destructuring to a super field is not supported yet."
* fix tests and fix assignment pattern
* remove CallExpression from AssignmentPattern
* test case for insertBefore for jsx
* fix unshiftContainer and insertBefore
* use path.scope.push
* add test making sure computedKeys var declaration at the right block
* add comment
* nit [skip ci]
### `@babel/runtime`
- Added `@babel/runtime-corejs3` package and `corejs: 3` options to `@babel/plugin-transform-runtime`.
- Added support of instance methods, fixes#8928.
- Added flag `proposals` (in `corejs: { version: 3, proposals: true }` format) for support all proposals polyfills from `core-js`.
- Used separate directories in runtime for `core-js` entry points with proposals and without.
- Used `get-iterator-method` helper for getting iterators, fixes#2500.
- As a cheap bonus, added support of IE8- (except some cases of `regenerator`).
### `@babel/polyfill`
- Should be deprecated in favor of separate usage required features from `core-js` and `regenerator-runtime` with an informative message.
### `@babel/preset-env`
- Uses for built-ins data from [`core-js-compat`](https://github.com/zloirock/core-js/tree/master/packages/core-js-compat) instead of `compat-table` since information from `compat-table` [is not enough](https://github.com/zloirock/core-js/tree/master/packages/core-js-compat).
- `useBuilIns` now requires direct setting of `corejs` version option, without it will be used `2` by default and shown deprecation warning.
- Added support of minor `core-js` versions for simplify updating in the future.
- For preventing some order-related problems, polyfills in the both `core-js@3` plugins added on `post` stage in the order of `core-js-compat` data.
- Divided plugins and polyfills parts of `preset-env`, instead of 2 internal plugins for adding polyfills, we have 6: usage and entry versions of plugins for `core-js@2`, ### Current state:
`core-js@3`, `regenerator-runtime`.
- Added support `samsung` target (for Samsung Internet) since `core-js-compat` and `compat-table` now contains mapping for this, fixes#6602.
#### `useBuilIns: entry` with `corejs: 3`
- No longer transforms `@babel/polyfill`.
- Transforms **all possible** `core-js` entry points to import of related modules (based on data from [`core-js-compat`](https://unpkg.com/core-js-compat@3.0.0-beta.15/entries.json)).
- Since of this, we no longer need `shippedProposals` / `proposals` flags with `useBuilIns: entry`.
- Removes `regenerator-runtime/runtime` import where it's not required.
#### `useBuilIns: usage` with `corejs: 3`
- In addition to `shippedProposals`, added flag `proposals` (in `corejs: { version: 3, proposals: true }` format) for polyfill all proposals from `core-js`.
- Fixed list of dependencies in built-in definitions.
- Improved the way of determination method / built-in name and source of this method.
- Adds import of required polyfills on `MemberExpression`, `ObjectPattern`, `in` operator.
- Adds import of required polyfills on access to global object properties.
- Adds import of all required common iterators on all syntax features which use iterators protocol (`for-of`, destructuring, spread, `yield` delegation, etc.).
- Adds import of promises on syntax features which use promises (async functions/generators, dynamic import, etc.), fixes#9250, #7402, etc.
### `core-js@2` stuff
I didn't want to tough `core-js@2`-related stuff, however
- Fixed some serious errors in definitions which breaks `Object.getOwnPropertySymbols`, `Symbol.toStringTag` logic, `Promise#finally`, `Array#forEach`, etc.
- `Array#flatMap` and trim methods moved to stable features as a part of ES2019 and loaded by deprecated `@babel/polyfill` and `@babel/preset-env` with `corejs: 2` option.
This is similar to ec69b4bb1256c061ac76f53dfed09c4283ec6a31, which
was about private instance fields.
Private properties can be non-writable (thanks to decorators), or have
get/set accessors. If we stored this information on the `privateClass`
object, we would need to always use `Object.getOwnPropertyDescriptor`
before reading or writing a property because accessors need to be called
with the correct `this` context (it should be the actual class, not the
object hat stores the private properties). This commit simplifies that
operation a bit by removing the container object.
It also have another advantage, which instance fields already have
thanks to the use of separate weakmaps: unused private static fields
can be tree-shaken away or garbage-collected, while properties of an
object can't. Also, they can be easilier minified.
* Private Static Class Fields Implementation
* Redo testing without class transform for static private
* Add a call() test for private static
* Use Object.defineProperty in loose mode
* Save full descriptor instead of only value for private fields.
Decorators can make private fields non-writable, so we need to store this
information somewhere.
The descriptor can also be used to implement private accessors.
This is going to be required so we can add support for other proposals,
as well as later set the accepted proposal as the default.
Update stage-0 and stage-1 presets with `pipelineProposal` to thread down to
the plugin.
* Revert "Move subclass inheritance to end (#7772)"
This reverts commit f8ab9466d331871a90f458af40b14e8d831e0c29.
* Only use getPrototypeOf if setPrototypeOf is implemented
* Update fixtures
* Helpers updates
* Update fixtures
* Fall back to getPrototypeOf
* Update fixtures
* Update member-expression-to-functions
1. Babel using British spellings, so `memoise`
2. Provide a helper `AssignmentMemoiser` class, which will assign the memo'd value with the `n`th access.
* Private properties!
* Fixes
* Tests
* Update helper name
* Fix privates that reference other privates
* Don't extend a builtin
* Rebase
This undoes the property call folding from #6656.
It complicates the private property transforms, since they boil down to `map.set(this, vlaue)` and we definitely don't want the next call define a property on the map.
/cc @Andarist
Yes, the output is uglier. But, this is necessary for me to refactor
`replaceSupers` for #7733, which is necessary for both #7555 and
https://github.com/babel/babel/pull/7553#issuecomment-381434519.
I'm still in the middle of cleaning up all this code. Don't expect
`transformClass` to survive much longer as it's written currently.
We were using `Object.create` to setup the prototype chain at the start of the class definition, which lead to #7771.
I was a bit worried about a speed hit, but it seems everyone optimizes the two patterns the same way.
https://jsbench.github.io/#f9fca52407643d96458a35763b201215Fixes#7771.
* Fix class properties after nested class' bare super
Fixes#7371.
* Fix node 4 test
* This damn node 4 test
* All of the ClassBody, but not the methods or field inits
* tmp
* tmp
* Use common class environment visitor
* Tests
* Use skipKey to avoid recursive traversal
* Remove old state
* Use jest expect
* Properly traverse nested class for supers
* Add object nested in class cases
* Add object nested object cases
* Test class properties
* Undo changes to lerna.json
* Add tests arournd prefix/postfix super increment
* tmp
* Use sets
* Improve get/set helper
* fixtures
* Edge cases
* Add loose edge cases
* Spec compliant
* Add issue case
* Even more edge cases!
* Final updates
* Fix name
* Use Reflect.{get, set} when available
* Avoid block scoping in loose
* Remove semicolon
* Do not redefine a non-enumerable
* Get strictness from call site, not helpers
* Add called assertions
* Classes are always strict
* Update test fixture
* #7349
- Add "use strict" directive to transformed class bodies
* Update tests (#7349)
* Add test for program with use strict directive (#7349)
* Update tests (#7349)
* Update source mapping (#7349)
* Add test for parent blockStatement with use strict directive (#7349)
* Update tests (#7349)
* Update tests (#7349)
* Update tests (#7349)
* Add test for constructor-only class (#7349)
* Constructor only classes are strict, too
But constructor only that use non-simple parameters must use a strict function wrapper.
* Fix test
* Wrapper not needed if class is already strict
* Revert change to lerna.json
It was needed to avoid collision of identifiers used in fields
initializers with variables declared in the constructor, but it
broke `this` handling.
This commit removes that behavior and instead renames the
colliding variables.
