**Description:**
This adds the ability to not include `ahash` with swc_common, which
caused some issues for me compiling dprint-plugin-typescript to Wasm
because of:
```
Compiling ahash v0.8.3
Compiling getrandom v0.2.10
error: the wasm*-unknown-unknown targets are not supported by default, you may need to enable the "js" feature. For more information see: https://docs.rs/getrandom/#webassembly-support
--> C:\Users\david\.cargo\registry\src\index.crates.io-6f17d22bba15001f\getrandom-0.2.10\src\lib.rs:285:9
|
285 | / compile_error!("the wasm*-unknown-unknown targets are not supported by \
286 | | default, you may need to enable the \"js\" feature. \
287 | | For more information see: \
288 | | https://docs.rs/getrandom/#webassembly-support");
| |________________________________________________________________________^
```
(I can't enable the JS feature because it's running the wasm file in
Wasmer and also I don't support Wasi in dprint plugins)
**BREAKING CHANGE:**
This removes swc_common's "perf" feature and makes it the default, then
adds an `ahash` feature instead. An alternative would be to make the
`ahash` dep optional and part of the default features, then do
`default-features = false` in the downstream crates (I think, but I'm
not sure), but I figure most people will be using the perf default
anyway? I'm not sure what's preferable.
**Related issue:**
- Closes#7729.
**Description:**
For crates using both `rkyv` and `swc`, where ast nodes from
`swc_ecma_ast` are included in data that's serialized with `rkyv`, it's
difficult to access the archived data from the archived ast nodes since
the types are private.
More context in the issue.
**Related issue:**
- Closes#7810.
bump() has a safety invariant that there must be input in the buffer to
read (See https://github.com/swc-project/swc/issues/7709).
This function as currently written calls `bump()` after calling
`read_token_interpreter()`, which *may* have exhausted the buffer
already in the specific case that it is handling a shebang. (The input
string `#!/bin/js` with no newline is sufficient to cause UB here)
Fortunately, shebangs never reach this function, due to `read_shebang`!
Hurray!
It's cleaner to not attempt to handle shebangs here when this code path
will never be reached, and use a debug assertion to ensure that we're
not dealing with shebangs.
**Description:**
The algorithm here is directly copied from terser, I don't if it's correct or can be improved, but it does fix the issue.
**Related issue:**
- Closes#7749
This PR replaces the current usage of lexical within the swc_ecma_parser
crate with equivalent parsing of large numbers using BigInt.
**Description:**
As discussed in
https://github.com/swc-project/swc/issues/7752, lexical contains a
number of soundness issues but doesn't appear to be actively supported.
Given the relatively low integration surface it seems reasonable to
replace the usage of lexical with another package to avoid this issue.
**Related issue:**
- Closes#7752
