fix: noUndeclaredVariables no longer errors on this in JSX tags (attempt 2)

[printfn]

Summary

Attempt 2 at fixing #2636. Relates to #2647.

I've added JsThisExpression as a new child node of AnyJsxElementName/AnyJsxObjectName, and changed the parser to no longer call p.re_lex(JsReLexContext::JsxIdentifier); when encountering a THIS_KW.

@Conaclos

Test Plan

I've added additional tests. It now seems to be correctly turning <this.foo /> into:

                        name: JsxMemberName {
                            object: JsThisExpression {
                                this_token: THIS_KW@65..69 "this" [] [],
                            },
                            dot_token: DOT@69..70 "." [] [],
                            member: JsName {
                                value_token: IDENT@70..73 "foo" [] [],
                            },
                        },

[codspeed-hq]

CodSpeed Performance Report

Merging #2656 will not alter performance

_{Comparing printfn:fix-jsx-this-bogus-nodes (9df96ee) with main (520cef0)}

Summary

✅ 84 untouched benchmarks

[Conaclos]

Thanks for the quick PR!

I need some time to think about that: should we add a specific node JsxThis or choose the approach you took, ...

[printfn]

I was thinking that it might make sense to try and stick closer to what the JSX gets transformed to.

Input:

<foo />;
<this.foo />;
<foo.bar />;

New JSX transform:

import { jsx as _jsx } from "react/jsx-runtime";
_jsx("foo", {});
_jsx(this.foo, {});
_jsx(foo.bar, {});

Old JSX transform:

import React from 'react';
React.createElement("foo", null);
React.createElement(this.foo, null);
React.createElement(foo.bar, null);

So I think it'd make sense to represent it similar to how this.foo etc. gets represented in normal JS. But there might be other considerations here that I'm missing.

[Conaclos]

I need some time to think about that: should we add a specific node JsxThis or choose the approach you took, ...

@denbezrukov Any opinion?

[ematipico]

I think we should keep the CST the same. I checked with Babel, and they do nothing in particular here. Instead, maybe, we should make an exception in our semantic model.

[Conaclos]

Closing in favor of #2636