This looks pretty cool! I think I looked at doing something like this a few years ago but, it got back-burnered in favor of other things.

Would the repr also include some of the other qualifiers like "set_name" or "set_parse_action"? Or just reconstruct the bare expression?

Yes, a PR to implement this would be very welcome.

I wasn't planning to show too much more than the required constructor arguments, lest it become unreadable too quickly. (The eval line above was meant only as a cute illustration, not as a proposed alternative to a class-aware copy() method.)

I'll polish something up for a PR.

PR is up, with a couple of questions:

1. _PendingSkip makes perfect sense in an And context, but I don't understand what it means to create one with the match-first | operator. (I wasn't able to find it in the documentation or figure it out from the unit tests.) What is X | Y | ... supposed to mean?
2. The Literal instances in an expression like Opt("A") + "B" are shown explicitly: Opt(Literal('A')) + Literal('B'). However, there are a number of contexts where the class name could be omitted. Cons: it would require introducing a second repr-like method to distinguish when you can and can't abbreviate, and ideally it should consider the current _literalStringClass. Pros: more concise output, and most classes would only need to implement either the regular __repr__() or the alternative.

With this change, what is the repr for Word(printables)?

Perhaps in the case of Word, we might use srange and _collapse_string_to_ranges when the initChars or bodyChars are very long strings.

The repr for printables is

!"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuvwxyz{|}~

so the repr for Word(printables) is nothing unexpected:

Word('!"#$%&\'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuvwxyz{|}~')

Using srange sounds reasonable anywhere we are dealing with a set of characters. (That said, shouldn't it return a set[str]?)

With the latest commit in the PR, the repr() for elements which are based on sets of characters now determines whether the character string or an srange call will take less space:

>>> Word(printables)
Word(srange('[!-~]'))
>>> Word("abcdefghijkl")
Word('abcdefghijkl')
>>> Word("abcdefghijklm")
Word('abcdefghijklm')
>>> Word("abcdefghijklmn")
Word(srange('[a-n]'))

I forgot to check out _collapse_string_to_ranges, but I'll see if that and the abbrev_charset functions I wrote can be harmonized.