formatter should translate HTML entities to code points where possible

[bakkot]

cf tc39/ecma262#404 (comment)

See also #476.

Note to self: this should not include blank codepoints like   or ‎. "blank" probably means "Control, White_Space, or Default_Ignorable", in Unicode terms. (Linked page does not include "Control" but it seems an obvious thing to add; it is not a subset of DI, per its definition, and indeed \u0000 is not DI.)

Also there's 93 entities which expand to multiple code points, often including a variation selector or "U+0338 COMBINING LONG SOLIDUS OVERLAY". Should exclude any multi-code point ligatures where at least one code point is blank in the above sense, and probably any which include combining characters (i.e. gc=M).

I.e. exclude anything which expands to a sequence of code points which matches /\p{White_Space}|\p{DI}|\p{gc=M}|\p{gc=C}/u.

[ljharb]

Additionally, can the linter require the formatter's output in this case?

[bakkot]

There's a --check flag to the formatter which asserts that the input is already formatted in the form which would be output by the formatter, which is intended to serve that purpose. (We use that in CI on 262.)