Think about dbColumnInfo() and dbFetch() output for unnamed columns

[krlmlr]

Should probably have NA in the "name" column, currently it's "NULL" or "?column?".

DBItest: "column_info".

[hadley]

What does an unnamed column mean in SQL? Oh you mean like a computed column?

[krlmlr]

The result of SELECT 1 UNION SELECT 2 is a two-row table with an unnamed column. It's valid SQL and can occur, and we should be able to handle it reasonably, also in dbFetch() and dbGetQuery().

[hadley]

SELECT 1 + 2 is an even simpler reprex. Agreed that NA most is reasonable here.

[krlmlr]

Not sure anymore: What should dbFetch() return for SELECT 1, 2? Data frames with two columns named "" don't seem to work well. Perhaps we should mandate a naming scheme for unnamed columns (e.g., implement a helper in DBI), and use this scheme also in dbColumnInfo().

[hadley]

Hmmm, maybe data.frame(X1 = 1, X2 = 2) ? So need to fill in missing variable names with auto generated in both places.

[krlmlr]

The data frame should have unique names, even for queries like SELECT 1, 1. For SQLite, we can determine if it's a computed column with column_database_name(), but I'm not sure about other backends. I think we're on the safer side if we don't specify that computed columns need to have artificial names.

[krlmlr]

Closing in favor of r-dbi/DBItest#137.