a better way to set the key in complicated one-liners

[eantonya]

If one tries to set the key on an intermediate expression before continuing, the flow breaks down, as the setkey has to be in front and the whole thing becomes a mess. Here's a random example (that doesn't do anything, and can be done differently, but imagine more complex stuff instead, e.g. http://stackoverflow.com/a/25628423/817778) of what I have in mind:

dt = data.table(a = 1:6, b = 1:2)

setkey(dt[, a[1], by = b], V1)[, c := 2*V1]

For a short period of time someone pointed out (I forget who, sorry) that it can be done like so:

dt[, a[1], by = b][, setkey(.SD, V1)][, c := 2*V1]

which I actually thought was fairly neat, but then this stopped being an option after .SD bindings got locked down.

I think we need to either allow the above .SD expression or have another alternative that can be written/read in a similar manner, from left to right, without having to skip back to the beginning of the sentence.

[arunsrinivasan]

.SD is allocated just once with the length of the maximum group, to avoid creating it again and again for each group. So I wouldn't opt for unlocking the binding, as it could lead to pretty nasty-to-track bugs.

[jangorecki]

You should still be able to use setattr(setorderv(.SD,cols),"sorted",cols), but I'm not sure how safe it is.

[MichaelChirico]

It's not clear to me from the example why we need to set the key here.

I found myself doing the messy mid-line setkey much more before on was an option; is this still a necessary feature? I'm struggling to think of an example where it's necessary, maybe something like:

setkey(dt[ , a[1], by = b], V1)[.(V1[1]), sum(a)]

Can't say I've ever needed something like that.

Also just recalled @jangorecki 's advice on #590:

If you use chaining you can also use fread(...)[, .SD,, "key"] which makes it a start point of chain for operations on sorted data.

[eantonya]

Agreed, on solves the issue neatly. Closing.