Type | Bug | Status | tested | Date | 1-Jul-2010 12:36 |
---|---|---|---|---|---|
Version | alpha 99 | Category | Mezzanine | Submitted by | BrianH |
Platform | All | Severity | minor | Priority | normal |
Summary | MOVE/to/skip has the wrong math for the destination index |
---|---|
Description |
MOVE/skip is supposed to scale the offset and length of the move by a record size, and it does. But when you specify MOVE/to, making the offset an index, it uses the wrong math to calculate the scaling: a simple multiplication. It should instead be a spanned index. Note: This also affects R2 2.7.7 and R2/Forward 2.100.80.1. |
Example code |
>> head move/skip at [a 1 2 b 3 4] 4 -1 3 == [b 3 4 a 1 2] ; correct >> head move/to/skip at [a 1 2 b 3 4] 4 1 3 == [a 1 b 3 4 2] ; should be [b 3 4 a 1 2] ; Note that the 1 was multiplied by 3 to get the destination index |
Assigned to | n/a | Fixed in | alpha 108 | Last Update | 20-Oct-2010 03:55 |
---|
Comments | |
---|---|
(0002418)
BrianH 1-Jul-2010 13:01 |
Fixed in R3 mezz-series.r 7551, R2 mezz-series.r 7552 and R2/Forward 2.100.80.2. |
(0002420)
Steeve 2-Jul-2010 10:37 |
I like the idea of the move function, but I just think it would be faster (and memory safe) to use the native swap in R3 to do such.
Though, currently we only can swap one value at a time. It would be convenient to add a looping behavior thru a refinement /size or /length. swap/size serie-1 serie-2 10 A move would be implemented by refering the same serie, 2 times. swap/size serie-1 skip serie-1 10 10 |
(0002421)
BrianH 2-Jul-2010 11:10 |
Your first example would work, but swapping a span of elements within the same series quickly runs into issues with overlapping spans. Part of the problem is that two series references are used, and halfway through the swap one of those references will become invalid, and remain invalid afterwards, as will the resulting data.
The whole reason that MOVE is included in R3 and R2 in the first place is that it is immune to aliasing and overlap problems, even before this fix is applied. Part of what makes MOVE safe to use is that it only has one series reference, and the rest are done by integer offsets, indexes and lengths. There was a long discussion that ended up with the MOVE function, and it was proven that these constraints are necessary to make the function safe to use. We could make MOVE native, but it would still need to behave the same. MOVE is there so that people won't be as tempted to reinvent it, since most of the time they end up with a function that doesn't work in some circumstances. It's a common problem. |
Date | User | Field | Action | Change |
---|---|---|---|---|
20-Oct-2010 03:55 | BrianH | Status | Modified | built => tested |
22-Sep-2010 06:36 | carl | Fixedin | Modified | => alpha 108 |
22-Sep-2010 06:36 | carl | Status | Modified | pending => built |
2-Jul-2010 11:23 | BrianH | Comment : 0002421 | Modified | - |
2-Jul-2010 11:20 | BrianH | Comment : 0002421 | Modified | - |
2-Jul-2010 11:18 | BrianH | Comment : 0002421 | Modified | - |
2-Jul-2010 11:11 | BrianH | Comment : 0002421 | Modified | - |
2-Jul-2010 11:10 | BrianH | Comment : 0002421 | Added | - |
2-Jul-2010 10:37 | Steeve | Comment : 0002420 | Added | - |
1-Jul-2010 13:02 | BrianH | Comment : 0002418 | Modified | - |
1-Jul-2010 13:02 | BrianH | Description | Modified | - |
1-Jul-2010 13:01 | BrianH | Comment : 0002418 | Added | - |
1-Jul-2010 13:00 | BrianH | Code | Modified | - |
1-Jul-2010 13:00 | BrianH | Status | Modified | submitted => pending |
1-Jul-2010 12:36 | BrianH | Ticket | Added | - |