Avoid "Inlined without wrapping"

[bergus]

The Expressions table does allow arbitrary values to be inlined "without inlining" in some contexts. I think that's a bad idea. This wording seems to suggest that the following things are allowed:

RegExp.make `a{${"1,2"}}` // /a{1,2}/
RegExp.make `a{1,${/2/}}` // /a{1,2}/
RegExp.make `a{${/b/}}` // /a{b}/ - matches "a{b}" literally
RegExp.make `a{${"1}b{2"}}` // /a{1}b{2}/
RegExp.make `a{${""}}` // /a{}/ - matches "a{}" literally

Urgh. Can we please do ToNumber on the interpolation value in the context of DecimalDigits, and throw a TypeError if it yields NaN?
So that the only allowed patterns are basically

RegExp.make `…{${repeat}}`
RegExp.make `…{${min},}`
RegExp.make `…{${min},${max}}`

where each of those variables has to be a number (or at least numeric).

[mikesamuel]

Should

RegExp.make`x{3,${ Infinity }}`

produce /x{3,}/

What about

RegExp.make`a{${ [1,2] }}`

[bergus]

@mikesamuel

Should RegExp.makex{3,${ Infinity }}`` produce /x{3,}/?

Sure, seems reasonable. That's what the Quantifier evaluates back to anyway…

What about RegExp.makea{${ [1,2] }}``?

No, that's not a number (and can't be coerced to one, like ["1"] could :-/). I don't see a reason to accept arrays. What if they're empty? What if they're longer than 2? If you are (ab)using arrrays as tuples and know that it has two elements, then it's not much harder to write

RegExp.make `a{${t[0]},${t[1]}}`

(Convince me that you need arrays here!)