Test that methods don't have special handling for row names by default

[krlmlr]

Automatic conversion of "row_names" column to row names.

• dbFetch()
• dbGetQuery()
• dbListFields()

[krlmlr]

Only for dbReadTable() and dbWriteTable(); the other methods should avoid touching row names by default.

Existing backends will need to make the default behavior configurable via pkgconfig.

[krlmlr]

@hadley @gaborcsardi: I'd like to specify dbGetQuery() to not convert a row_names column to row names by default -- only dbReadTable() and dbWriteTable() should do this. Do you anticipate any problems here (except backward compatibility for existing backends, which can be handled with pkgconfig)?

[hadley]

Why do you want the behaviour to be different between these functions?

[gaborcsardi]

@hadley seconded. Unless it is causing trouble, I would not change it.

[krlmlr]

They operate at different conceptual levels.

dbReadTable() and dbWriteTable() attempt to serialize and deserialize a data frame. It's okay to support row names there, but we may as well draw the line here and not support them at all, at least not by default. (Again, pkgconfig.)

dbGetQuery() is about getting data from arbitrary queries into R. Attaching special semantics to a column of a particular name is difficult to justify IMO, and calls for trouble. Just two examples:

• Some implementations of dbListFields() omit the row_names column
• dplyr doesn't roundtrip it: memdb_frame(row_names = 1:3)

I'd prefer to get rid of the default row names conversion for dbGetQuery(), and possibly also for dbReadTable() and dbWriteTable().

[hadley]

Didn't we try and get rid of it for dbReadTable() in the past and it caused a lot of problems? In an ideal world, there never would have been special handling for row names, but given how established DBI is, I think it's going to be hard to change now.

[krlmlr]

row.names = FALSE is now the default for RSQLite, with a pkgconfig switch. Let's see how this plays out.