cc @mathiasbynens

I agree the current paragraph is oddly phrased. I don't understand why it's written like that. Maybe @NorbertLindenberg or @allenwb remember?

If we're gonna change this, your first suggested approach seems simplest IMHO:

If some part of the source text matches \u LeadSurrogate \u TrailSurrogate, it must be parsed as a single Atom rather than two.

Note that PR #1727 also addresses this issue.

I didn’t write this – it first showed up in the spec draft of 2015-04-03. The change was preceded by a discussion on “lonely surrogates and unicode regexps” starting here:
https://mail.mozilla.org/pipermail/es-discuss/2015-January/041281.html

I wrote it because without it (or something like it) the grammar is ambiguous. The reason the phrasing parallels the language used for "dangling else" clause is because at the grammar engineering level this is the same kind of ambiguity and copying that language seemed like a safe choice to make given this was happening in April 2015 immediately before release of the final ES6 draft. There wasn't a lot of time to study alternatives.

Note that the disambiguation really only concerns the first three alternative of

RegExpUnicodeEscapeSequence ::
u LeadSurrogate \u TrailSurrogate
u LeadSurrogate
u TailSurrogate
...

because that is where the ambiguity exists. The disambiguation statement does not lead to recognizing "nested surrogate pairs" in situations such as such as:

\u LeadSurrogate \u LeadSurrogate \u TrailSurrogate \u TrailSurrogate

1. In processing the first lead surrogate there is no ambiguity between choosing between alternative 1 and alternative 2 because because the lead surrogate is not immediately followed by a tail surrogate. So, alternative 1 is matched.
2. In processing the second surrogate there is an ambiguity. Either alternative 1 or alternative 2 could be used. So, the disambiguation rule applies and forces use of alternative 1, consuming the first tail surrogate.
3. Finally, the second tail surrogate is processed unambiguously matched using alternative 3.

So, no "nesting" is recognized.

But, I agree that the current wording of the disambiguation statement isn't ideal for this situation. My suggestion for a replacement (I excluded the grammar parameters but the actual spec text should include them):

For ambiguous inputs where either the rule:
RegExpUnicodeEscapeSequence :: u LeadSurrogate \u TrailSurrogate
or the rule:
RegExpUnicodeEscapeSequence :: u LeadSurrogate
could be recognized the first rule always is the one that is recognized.

There's another problem with that paragraph: it makes it sound like there can be any number of non-surrogate, and even non-\u characters between the lead and trail surrogates, which I don't think should be allowed.

Attached is the email thread that led to the addition of the paragraph in question.
Note that it had nothing to do with nesting of surrogate pairs. Its only intent was to disambiguate the grammar.

Re ES6 grammar abnormalities.pdf