Grammar parses != in right-associative way

[f-f]

According to the standard all operators are left-associative

For example, the following expression:

True == False != False

should parse in this way:

(True == False) != False

-- Or in CBOR encoding:
-- [3 3 [3 2 true false] false]

However, the only parse the grammar allows in this case is the following:

True == (False != False)

-- Or in CBOR encoding:
-- [3 2 true [3 3 false false]]

The issue should be here, as an "equal expression" can contain "not-equal-expression"s, but not viceversa 🤔

[Gabriella439]

My understanding is that associativity only specifies how multiple occurrences of the same operator are parenthesized. If you combine different operators then operator precedence governs how they are parenthesized and that takes precedence (pun intended) over associativity.

[f-f]

Oh right, makes sense. Then I guess the grammar is also the proper reference for operator precedence (as it limits the possible parses one can get, as in this case)

But then it looks like the parsing by the Haskell implementation not using the right operator precedence: in the parser/success/operators test the True == False != False is parsed as (True == False) != False, which is not allowed by the grammar

[Gabriella439]

Oh, that's a bug in the Haskell implementation. It has the precedence of the != and == operators backwards:

https://github.com/dhall-lang/dhall-haskell/blob/8663bfe2967fbad07590a17e90abe60c4ab0c170/dhall/src/Dhall/Parser/Expression.hs#L175-L181

I can fix that

[Gabriella439]

Fix is up here: dhall-lang/dhall-haskell#825