Alternative interpretation of -strides option

[Lestropie]

Follows from #3271; not going to do it there to expedite completion since it's not necessary to replicate existing behaviour.

There are actually a couple of different ways in which the -strides option may be used. In the parlance of #3271, sometimes it is used to specify something more akin to a permutation (eg. 0,0,0,1 for volume-contiguous; though "0" here is an ignore flag, an actual permutation for this purpose would be 1,1,1,0), sometimes it is used to specify symbolic strides (eg. -1,+2,+3,+4).

I think these two cases could in fact be separated by applying the following categorisation to each value:

1. Explicit negative value
2. Explicit positive value (ie. "+" present)
3. Implicitly positive due to absence of sign symbol

Perhaps 1 and 2 should be treated as a symbolic stride with sign, whereas 3 should be treated only as a magnitude, preserving the sign of the input image if advantageous to do so?

MR::parse_ints<>() already parses the presence / absence of "-" and "+". So this information could be accessed by having MR::parse_ints<>() call some other function that does the heavy-lifting but returns both values and the categories above, but discards the latter. The parsing of specifically -strides could then invoke that function directly in order to preserve the implicit / explicit-ness of the signs.

[kikiluvbrains]

One question, How would mixed inputs be handled? For example something say like -1,2,+3 would that be treated partially as symbolic and partially as magnitude only?

[Lestropie]

Yes, the implicit vs. explicit nature of the sign of each entry would need to be tracked individually; the code that resolves the input image strides against the command-line user request would then choose whether to override the sign of the symbolic stride of each axis individually.