This pr introduces some cargo features. For `swc_common` and `swc_ecma_ast`, it introduces a feature flag `fold`.
`Fold` and `Visit` traits exist only if the feature is enabled.
For `swc_ecma_parser`, flag called `verify` is added. When disabled, we skip checking of validity of some expressions.
e.g. `{foo = bar}`
Verification is disabled by default it requires nightly compiler
ecma_transforms:
- implement es2015::instanceof
- implement es2015::typeof_symbol
- implement inline_globals pass
ecma_parser:
- `PResult<T>` is now `Result<T, ()>` and `Err(())` means that an error is emitted.
- add docs
- rename packages to be consistent
- `swc_macros` is removed. Now macros are imported with `extern crate macro_name` instead of `extern crate swc_macros`.
- manage atoms with words.txt file
* `struct Expr` + `enum ExprKind` -> `enum Expr`
`Expr` / `ExprKind` approach does not work well with recursive
processing based on the type system because we can't access common
fields like span while processing child node like `MemberExpr`.
As ast processing is inherently recursive, we should use a
recursive-friendly approach.
* `#[derive(FromVariants)]`
It creates lots of `From<T>` to help using ast enums as a real sum type.
* `#[derive(Spanned)]`
Inspired by https://docs.rs/syn/0.12.13/syn/spanned/trait.Spanned.html
It does not handle attributes yet.
* Parser is not updated yet.
rust-analysis chokes with `Option<Box<Expr>>` style fields.
Rls only shows a warning on the crate with that fields, but ICEs on
dependent crates.
This can be workarounded by using `Option<(Box<Expr>)>`.
Also, using multiple glob imports is bad for rls.
So this commit deglobs them.
- Parser and lexer for lastest ecma spec https://tc39.github.io/ecma262
- Lexer is currently very inefficient
- Use https://github.com/tc39/test262-parser-tests/ for testing.
- Implement proc-macro based ast folder and assert_eq_ignore_span! based on it.
- Some utilities for proc macro at /macros/common
