> Then documentation should say that, not only scalar variables with names '\', '1' are valid, but also these names are valid for all structures.

I agree.

> Otherwise it is a bug.

It's a feature! ;]

Honestly there are many issues with these "open source" docs, and I don't know how to solve them.

The usual response is "contribute", but the result of more authors is just more confusion.

We are somehow at the limit of the possibilities of self managed groups. (This observation doesn't only apply for the docs)

regarding special symbols

though in this particular case I'd suggest adding a remark to the beginning of perlvar and/or a caveat section at the end, to limit the amount of confusion.