Precedence of transposition operator '

[nalimilan]

Though not terrible, this is a bit unfortunate:

julia> :(1:4')
:(1:(4)')

I can't find a case where one could want to create a range from a transposed last element. OTC, creating a row-matrix from a range can sometimes be useful.

Maybe we could change this? Or would it break something else?

Also, ' isn't listed in the table about operator precedence.

[stevengj]

Theoretically, this would be a breaking change, although it's hard to imagine it affecting real code.

[TotalVerb]

Currently transposition has higher precedence than multiplication and addition, and that's probably a good thing: A + B' should probably mean A + (B'), not (A + B)'.

But range has lower precedence than both: 1:10*10 is 1:(10*10). This change would have to break that behaviour. Especially concerning is this:

1:length(A)-1

which I think is likely quite common.

[timholy]

Seems like the easiest way to solve this is to make 4' an error. Currently ctranspose is recursive (to handle arrays-of-arrays) and on numbers it computes conj. But perhaps we could make ctranspose recurse to _ctranspose (non-exported), and restrict ctranspose to AbstractVecOrMat.

[stevengj]

@TotalVerb, good point, we can't make ' lower precedence than : without breaking the relationship with + (or breaking the partial ordering of precedences).

@timholy, please no. ' is adjoint, and making the adjoint of a number equal to its conjugate is extremely well established mathematics. It seems crazy to break that (making it harder to write generic code that works with both matrices and numbers) just to give a syntax error for odd mistakes with colons.

[nalimilan]

I don't think there really a problem here. I was just wondering whether we could make this slightly more convenient without making anything more painful.

I'm inclined to close this issue. But I am right that ' should be added to the operator precedence table? We can turn it into a docs issue.

[TotalVerb]

I think implicit multiplication (which binds quite strongly, but not as strong as exponentiation) is also missing from that table, along with probably many other common ones. All the unicode ones are missing too, but those can probably be left out.

[JeffBezanson]

This doesn't seem to be solvable with transitive operator precedence. : has lower precedence than +, but transpose has to have higher precedence than +. #18714 would be the change we could make here if anything.