prefer-string-raw: Add support for template literals

[som-sm]

Closes #2653

[fisker]

I'm a little worried about how \r works, but I'm going to pretend I don't know it. It's quite tricky in template literals.

[som-sm]

I'm a little worried about how \r works, but I'm going to pretend I don't know it. It's quite tricky in template literals.

Is this comment for this PR or the other one?

[fisker]

This one.

[som-sm]

This one.

Ok, I'm not sure if I follow, will \r behave any differently from other escaped characters?

If the string contains \r, then raw shouldn't be equal to cooked and it shouldn't complain.

[fisker]

> require('espree').parse('`\r\r\n`',{ecmaVersion:'latest'}).body[0].expression.quasis[0]
Node {
  type: 'TemplateElement',
  start: 0,
  end: 5,
  value: { raw: '\n\n', cooked: '\n\n' },
  tail: true
}
> require('@typescript-eslint/typescript-estree').parse('`\r\r\n`').body[0].expression.quasis[0]
{
  type: 'TemplateElement',
  tail: true,
  value: { cooked: '\n\n', raw: '\r\r\n' }
}
> require('@babel/parser').parse('`\r\r\n`').program.body[0].expression.quasis[0]
Node {
  type: 'TemplateElement',
  start: 1,
  end: 4,
  loc: SourceLocation {
    start: Position { line: 1, column: 1, index: 1 },
    end: Position { line: 3, column: 0, index: 4 },
    filename: undefined,
    identifierName: undefined
  },
  value: { raw: '\n\n', cooked: '\n\n' },
  tail: true
}

As far as I know, it's not really clear in ESTree spec.

[som-sm]

@fisker Looked into this a bit, here are my observations:

With espree and @babel/parser, there shouldn't be any issues since the raw and cooked values are the same.

The only thing is that in some situations we won't be preserving carriage returns, like this:

test.babel({
	valid: [],
	invalid: [
		// eslint-disable-next-line internal/no-test-only
		test.only({
			code: 'a = `a\\\\b \r\n`',
			output: 'a = String.raw`a\\b \r\n`',
			errors: 1,
		}),
	],
});

Is this a problem?

---

With the @typescript-eslint/typescript-estree parser, inputs like 'a = `a\\\\b \r\n`' should ideally be reported—but they won't, because the unescaped value wouldn't be equal to the cooked value. For example:

[
  {
    cooked: 'a\\b \n',
    raw: 'a\\\\b \r\n',
    unescapeBackslash: 'a\\b \r\n'
  }
]

Maybe we can fix this by replacing all \r and \r\n characters with \n in the raw value, like:

function normalizeCarriageReturn(text) {
	return text.replaceAll(/\r\n?/g, '\n');
}

if (
	(node.parent.type === 'TaggedTemplateExpression' && node.parent.quasi === node)
-	|| node.quasis.every(({value: {cooked, raw}}) => cooked === raw)
+	|| node.quasis.every(({value: {cooked, raw}}) => cooked === normalizeCarriageReturn(raw))
-	|| node.quasis.some(({value: {cooked, raw}}) => cooked.at(-1) === BACKSLASH || unescapeBackslash(raw) !== cooked)
+	|| node.quasis.some(({value: {cooked, raw}}) => cooked.at(-1) === BACKSLASH || unescapeBackslash(normalizeCarriageReturn(raw)) !== cooked)
) {
	return;
}

This would make cases like these be reported, but the carriage return would still not be preserved. So, you would get 'a = String.raw`a\\b \n`' for 'a = `a\\\\b \r\n`'

WDYT?

[fisker]

I prefer just ignore it, if user complain, we can just blame typescript-eslint 😄