Breaking change in 2.7.0: false positive 'expression inside repetition is non-progressing and will repeat infinitely'

[reivilibre]

Describe the bug Pest 2.7.0 seems to have introduced a breaking change that causes my library to fail to compile. (Previously I was using 2.5.5, I have confirmed 2.5.7 to work and cannot comment on 2.6.* because that series seems to have been yanked from the crate registry).

To Reproduce Steps to reproduce the behavior:

The following snippet illustrates the issue and can be tried on the pest homepage (I have cut out some unrelated productions):

nlOrEoi = _{ NEWLINE | &EOI }

ws = _{ " " | "\t" }
ws_nocnl = _{ " " | "\t" }

// Line ending, with allowed blank lines.
lineEnd = _{ ws_nocnl* ~ nlOrEoi ~ (!EOI ~ ws* ~ nlOrEoi)* }

This gives an error (taken from my real project, but the snippet is representative):

error: proc-macro derive panicked
  --> project/src/parser.rs:20:10
   |
20 | #[derive(Parser)]
   |          ^^^^^^
   |
   = help: message: grammar error
           
             --> 97:36
              |
           97 | lineEnd = _{ ws_nocnl* ~ nlOrEoi ~ (!EOI ~ ws* ~ nlOrEoi)* }
              |                                    ^---------------------^
              |
              = expression inside repetition is non-progressing and will repeat infinitely

Expected behavior

This expression is actually always progressing, though I can see why the compiler is unable to prove it — were it not for !EOI then I suspect it would be right. The compiler should not raise an error on this, particularly since it was working in a previous (semver-compatible) version of Pest.

Additional context

The grammar may look a bit unnecessarily convoluted and academic, but the real version deals with e.g. comments etc and has a bit more complexity to it. (Definitely not saying there isn't a simpler way of doing this, but I can't easily see one :))

[reivilibre]

Well, no need to fix it just for me, I found a clearer way to do what I wanted that doesn't cause this error:

lineEnd = _{ ws_nocnl* ~ nlOrEoi ~ (ws* ~ NEWLINE)* ~ (ws* ~ &EOI)? }

However maybe this is biting other people, so it's up to you :).

[tomtau]

@reivilibre I'm guessing it may be due to this PR: https://github.com/pest-parser/pest/pull/848/files that landed in 2.6.X I think @Tartasprint was trying to capture more cases: https://github.com/pest-parser/pest/discussions/851

@reivilibre do you think this is a false positive or should we convert this into Q&A just in case people run into that issue?

[Tartasprint]

I do not think this is easy to fix. It is easy to understand why the expression makes progress, but it requires an in-depth analysis for a program.

To make this possible I suggest (as in #851 ) making annotations to the AST of the Pest file being read at the verification stage.

I do want to make progress on this, but I won't be able to make any until mid-November.