Add index option to compute()

[iangow]

It would be great if collect() could be accompanied by the creation of indexes for performance.

[hadley]

Not sure how that would help - collect() bring all the data into R, where there aren't indices.

[iangow]

Sorry. I think I should've said compute().

But collect() is what I've ended up doing.

I am merging A with B to produce AB, which is bigger than either, then merging with C to produce something smaller. I figure a server-side merge of AB with C assisted by an index would be faster than bringing the data to R. But if I'm using RStudio Server, then it's no big deal for the data I'm working with. (If I were using RStudio with a remote database, then there'd be more of a difference.)

(Of course, if one could slip arbitrary SQL in through the dplyr-constructed connections, then one could make indexes "by hand".)

[hadley]

Ah yes, there should be an option to do that - it's easy to add.

[krlmlr]

Checking that a column is unique is another possible application. Perhaps compute() can be taught to create both unique and non-unique indexes. Mind that indexes can include multiple columns.