Unicode escape sequence in string literal is decoded incorrectly

[adams85]

What version of Bun is running?

1.1.36

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What steps can reproduce the bug?

repro.js:

const s = "\uFFFF";

console.log("length:", s.length);

for (let i = 0, ch; !isNaN(ch = s.charCodeAt(i)); i++) {
  console.log("\\u" + ch.toString(16).padStart(4, '0'));
}

bun run repro.js

What is the expected behavior?

length: 1
\uffff

What do you see instead?

length: 1
\udbff
\udfff

Additional information

According to my tests, 1.1.34 decodes the escape sequence correctly. So this is a regression introduced in 1.1.35 (it's present in 1.1.36 as well).

Both "\uFFFF" and "\u{FFFF}" are decoded incorrectly.

[Jarred-Sumner]

Transpiler output:

const s = "\uDBFF\uDFFF";
console.log("length:", 1);
for (let i = 0, ch;!isNaN(ch = "\uDBFF\uDFFF".charCodeAt(i)); i++)
  console.log("\\u" + ch.toString(16).padStart(4, "0"));

❯ bun-1.1.34 build --target=bun a.js
// @bun
// a.js
var s = "\uFFFF";
console.log("length:", s.length);
for (let i = 0, ch;!isNaN(ch = s.charCodeAt(i)); i++) {
  console.log("\\u" + ch.toString(16).padStart(4, "0"));
}

❯ bun build --target=bun a.js
// @bun
// a.js
var s = "\uDBFF\uDFFF";
console.log("length:", s.length);
for (let i = 0, ch;!isNaN(ch = s.charCodeAt(i)); i++) {
  console.log("\\u" + ch.toString(16).padStart(4, "0"));
}