Moving this issue and the corresponding PR to milestone 0.35 because it makes ';' ambiguous in the context of group aggregates. For example:

arr.sum('a2;b1..b3') 

currently means two groups on the same axis. We have two options here:

• either use another separator between groups on the same axis. This would be cleanest but would break backward compatibility. For example, in the following there would be a b1,b2,b3,b5 group and another b7,b8,b9 group:

arr.sum('a1,a3;b1..b3,b5/b7:b9') 

• somehow (unsure it is possible to implement this easily/without introducing some ambiguous cases) allow several groups from the same axis and groups from different axes using the same separator. For example, make this work:

arr.sum('a1,a3;b1..b3,b5;b7:b9') 

In that case, we could require the groups to be enclosed by [ ] or ( ) and separate groups on the same axis by , (but that would be backward incompatible)

arr.sum('a1,a3;[b1..b3],[b5:b7]') 

Either way, this needs more thoughts.

@alixdamman : if you have any idea on how to solve this, please speak 😉

In that case, we could require the groups to be enclosed by [ ] or ( ) and separate groups on the same axis by , (but that would be backward incompatible)

arr.sum('a1,a3;[b1..b3],[b5:b7]') 

Using [ and ] seems like a bad idea to me now because we cannot reproduce this outside the string syntax and it makes the [ kinda ambiguous.

Using ( and ) around the different groups on the same axis would make more sense I think (this would match pretty closely with what we need to do with the non-string syntax):

# match two groups -> group aggregate
arr.sum('a1,a3;(b1..b3,b5;b7:b9)') 
# a single group -> axis disappears
arr.sum('a1,a3;b1..b3,b5,b7:b9') 

In any cases (parentheses or not), we will need to decide if we want to support this:

# match two groups on b -> group aggregate
arr.sum('a1,a3;b1..b3,b5;b7:b9')