potentially confusing behavior in cartesian loops

[StefanKarpinski]

Not sure if this is a bug, but it did surprise me:

julia> for x = int64(2)^53-2:int64(2)^53+5, y=1:2
         println(x)
         x = uint64(x)
       end
9007199254740990
0x001ffffffffffffe
9007199254740991
0x001fffffffffffff
9007199254740992
0x0020000000000000
9007199254740993
0x0020000000000001
9007199254740994
0x0020000000000002
9007199254740995
0x0020000000000003
9007199254740996
0x0020000000000004
9007199254740997
0x0020000000000005

[vtjnash]

thought this might be a little clearer (if I understand the bug/unexpected behavior correctly)

julia> for x = 1:10, y=1:2
               println(x," ", y)
       end
1 1
1 2
2 1
2 2
3 1
3 2
4 1
4 2
5 1
5 2
6 1
6 2
7 1
7 2
8 1
8 2
9 1
9 2
10 1
10 2

[JeffBezanson]

I don't see what else this could do.

[StefanKarpinski]

@vtjnash: That's actually what it's supposed to do. It's a shorthand for a pair of nested for loops.

@JeffBezanson: I guess why I found it surprising is that we're used to having loop variables reset on each iteration, regardless of what we do with them. In this case, however, x iterations and y iterations are not the same, despite them appearing to be in the same "superloop". It really is just syntax for nested for loops. It could mean something like this instead:

for _x = int64(2)^53-2:int64(2)^53+5, _y=1:2
  x = _x; y = _y
  println(x)
  x = uint64(x)
end

[JeffBezanson]

My worry is that somebody might rewrite one form to the other, thinking it is just a syntax shortcut. So the behavior shouldn't change.

[StefanKarpinski]

Let's just leave this behavior as-is and close this issue.

[cmey]

This hit me today, I was doing something like this in image processing:

for x=1:4,y=1:2
       #processing using x,y
       println( (x,y) );
       (y,x) = (x,y)
       #processing using transposed x,y
end

The behavior of the cartesian loop syntax was very confusing.
Now I understand that if I had written two separate loops, that behavior would have been expected. But in the cartesian loop syntax, I expected the variables x and y to be reset, like @StefanKarpinski said.

[rfourquet]

This was closed in 2012, but in the meantime, the break behavior has changed from within "multiloops" (with one for keyword), in #7430. Is it worth reconsidering this issue?

[JeffBezanson]

Yes, that's a good point. Either option seems ok to me; I'll sleep on it. Would be interesting to see what others think.

[StefanKarpinski]

I can see it either way, but I still think my original example is a bit of a gotcha.

[rfourquet]

The comprehensions seem to suggest a change:

julia>  [let z=(x, y); x=0; z end for x=1:2 for y=1:2]
4-element Array{Tuple{Int64,Int64},1}:
 (1, 1)
 (0, 2)
 (2, 1)
 (0, 2)

julia>  [let z=(x, y); x=0; z end for x=1:2, y=1:2]
2×2 Array{Tuple{Int64,Int64},2}:
 (1, 1)  (1, 2)
 (2, 1)  (2, 2)