Literal NaNs in ASTs

[Keno]

Perhaps comparing ASTs isn't a very common operation, but I came across the following which I found surprising. I think it is even arguable that == should ignore the facts that NaNs are not numerically equal to themselves since it's not really numerical comparison.

julia> :(f($NaN)) == :(f($NaN))
false

julia> isequal(:(f($NaN)), :(f($NaN)))
false

[JeffBezanson]

It looks like we need an isequal method for Expr that calls isequal recursively.

[wizzard0]

it is even arguable that == should ignore the facts that NaNs are not numerically equal to themselves since it's not really numerical comparison.

NaN literal is not NaN value, yep

[JeffBezanson]

Very good point. The real solution is probably for == on Exprs to use isequal to compare elements.