fix: skip ascii symbol in process

[JLHwung]

In this PR we skip the processItem when it is a symbol-kind and has code point within ASCII range.

This fix will prevent - transpiled as \\x2D when it is not in character class. Besides that, it should speed up a little bit since ASCII symbols are pretty common.

[nicolo-ribaudo]

It would be good to have a test with - inside a character class, e.g. /[--\\]/

[JLHwung]

@nicolo-ribaudo Good question! This PR does not touch - in character class by design.

Currently, every items in CharacterClass is added to a single regenerate set so that we can perform possible code range concatenations. However, the regenerate lacks of the token information: i.e. the kind of token where - is a symbol and \- is an identifier. Therefore, regenerate has to output escaped \\x2D even if it may be redundant within the context.

[mathiasbynens]

Did you see #16?

[JLHwung]

Sorry I didn't see #16 before.

What's happening in #16 is that - is passed to regenerate and regenerate.toString returns \-, which is invalid when unicode flag is enabled, so regenerate now returns \x2D.

However, when - is a symbol-kind and not in a character class, it should not be transpiled at all. Note that \- is an identifier-kind, which means \- will still pass into regenerate and output \x2D.

[mathiasbynens]

I'm okay with making - not be escaped outside of character classes. It’s correct that this logic needs to live in regexpu-core and not regenerate, since regenerate’s output must work in any context (in and outside character classes).

But let’s keep the behavior for other ASCII symbols (in particular, non-printable ones) as-is.

[nicolo-ribaudo]

Let's keep this as codePoint == 0x2D, since many ascii symbols must be escaped: [, {, |, +, ?, *, $, ...

[JLHwung]

Escaped character, i.e. \[ is identifier kind instead of symbol kind. If regjsparser parses it safely as symbol, we don't have to escape it.

This is how /\{/ will be parsed:

{
  "type": "value",
  "kind": "identifier",
  "codePoint": 123,
  "range": [
    0,
    2
  ],
  "raw": "\\{"
}

[JLHwung]

The CI error is fixed in #81.

[nicolo-ribaudo]

@JLHwung could you rebase?