Examples for solving conflicts with negative predicates

[twolodzko]

Describe the bug

Negative predicate following a positive match of the same symbol seem to conflict with each other when parsing. This may be either bug or a case for documentation improvements if I'm understanding something incorrectly.

To Reproduce

For a trivial language:

expr = { term ~ (op ~ term)* }
term = { atom ~ ( bang ~ !"=" )? }
atom = _{ value | "(" ~ expr ~ ")" }
bang = { "!" }
op = { "+" | "-" | "=" | "!=" }
value = { ASCII_ALPHANUMERIC+ }
WHITESPACE = { " " }

I would expect something like x! = 2 to parse as value ~ bang ~ op ~ value, but it does not. I was trying to add positive predicates for whitespaces etc as well, but with no luck to make it work.

x != 2 and (x!)=2 and x!=2 parse correctly.

Expected behavior

x! = 2 parses correctly as value ~ bang ~ op ~ value.

[tomtau]

I'm not sure if it's a bug. You can try stepping through the parsing process to see what rules are being applied in a debugger, e.g. https://tomtau.github.io/pest-web-debug/
My hunch is that it's due eagerness + ordered choices + the implicit whitespaces. e.g. if I modify the first rule to force it to consume the whole input: expr = { term ~ (op ~ term)* ~ EOI }, it'll then flag =:

 --> 1:4
  |
1 | x! = 2
  |    ^---
  |
  = expected WHITESPACE

[tomtau]

https://github.com/pest-parser/awesome-pest/blob/main/debugger.md

[twolodzko]

I was not able to fix it in my grammar with the debugger. The grammar is a bit ambiguous, but I hoped to resolve it with the negative lookahead ~ !"=", but it does not seem to work. Is there other way in Pest to declare such grammar?

[tomtau]

How about this:

expr       = { term ~ (op ~ term)* }
term       = { atom ~ postfix_bang? }
postfix_bang = @{ "!" ~ !"=" }
atom       = _{ value | "(" ~ expr ~ ")" }
op         = { "!=" | "+" | "-" | "=" }
value      = { ASCII_ALPHANUMERIC+ }
WHITESPACE = _{ " " }
``` ?

[twolodzko]

Yes, it works! Maybe some example of solving conflicts like this could land in the book? It feels like there's much more to @{ ... } than ignoring whitespaces.

[tomtau]

Ok, I moved this issue under the book repo