Fixup/oracle dates

[detule]

Fixes long-standing ORACLE issues with using batch sizes greater than one when writing to tables with DATE and TIMESTAMP fields.

See, for example #349, #350, #391

[simonpcouch]

Unable to push this around locally before the upcoming release, but these changes feel like a step forward for me!

[simonpcouch]

Do we need to revert this column back to another type after processing it?

[detule]

Hey Simon:

Good question - I think the query can/does return all NULLs for that field, which in turn I think gives a column full of NA_character_ in the data-frame.

[hadley]

This is a very minor quibble but I don't trust this sort of cross-column subset-assignment, so I'd prefer to see something more like:

res$data_type[res$field.type == "DATE"] <- 91
res$column_size[res$field.type == "DATE"] <- 6

I suspect that might also eliminate the need for as.numeric(res$data_type) above.

[detule]

Thanks. Done - let me know if that's not what you had in mind.

I think the as.numeric may still be needed, but happy to update if you guys think there's a better way.

 ┃  Browse[1]> res <- dbGetQuery(conn, query)
 ┃  Browse[1]> str(res)
 ┃  'data.frame':   5 obs. of  17 variables:
 ┃   $ name                   : chr  "datetime" "date" "integer" "double" ...
 ┃   $ field.type             : chr  "TIMESTAMP(6)" "DATE" "NUMBER" "BINARY_DOUBLE" ...
 ┃   $ table_name             : chr  "test_table" "test_table" "test_table" "test_table" ...
 ┃   $ schema_name            : chr  "SA" "SA" "SA" "SA" ...
 ┃   $ catalog_name           : chr  NA NA NA NA ...
 ┃   $ data_type              : chr  NA NA NA NA ...
 ┃   $ column_size            : num  NA NA NA NA 255
 ┃   $ buffer_length          : num  11 16 40 8 255
 ┃   $ decimal_digits         : num  6 NA 0 NA NA
 ┃   $ numeric_precision_radix: chr  NA NA NA NA ...
 ┃   $ remarks                : chr  NA NA NA NA ...
 ┃   $ column_default         : chr  NA NA NA NA ...
 ┃   $ sql_data_type          : chr  NA NA NA NA ...
 ┃   $ sql_datetime_subtype   : chr  NA NA NA NA ...
 ┃   $ char_octet_length      : num  0 0 0 0 255
 ┃   $ ordinal_position       : num  1 2 3 4 5
 ┃   $ nullable               : num  1 1 1 1 1
 ┃  Browse[1]>     isDate <- res$field.type == "DATE"
 ┃      res$data_type[isDate] <- 91
 ┃      res$column_size[isDate] <- 6
 ┃      isTimestamp <- grepl("TIMESTAMP", res$field.type)
 ┃      res$data_type[isTimestamp] <- 93
 ┃      res$column_size[isTimestamp] <- 16
 ┃  Browse[1]> str(res)
 ┃  'data.frame':   5 obs. of  17 variables:
 ┃   $ name                   : chr  "datetime" "date" "integer" "double" ...
 ┃   $ field.type             : chr  "TIMESTAMP(6)" "DATE" "NUMBER" "BINARY_DOUBLE" ...
 ┃   $ table_name             : chr  "test_table" "test_table" "test_table" "test_table" ...
 ┃   $ schema_name            : chr  "SA" "SA" "SA" "SA" ...
 ┃   $ catalog_name           : chr  NA NA NA NA ...
 ┃   $ data_type              : chr  "93" "91" NA NA ...
 ┃   $ column_size            : num  16 6 NA NA 255
 ┃   $ buffer_length          : num  11 16 40 8 255
 ┃   $ decimal_digits         : num  6 NA 0 NA NA
 ┃   $ numeric_precision_radix: chr  NA NA NA NA ...
 ┃   $ remarks                : chr  NA NA NA NA ...
 ┃   $ column_default         : chr  NA NA NA NA ...
 ┃   $ sql_data_type          : chr  NA NA NA NA ...
 ┃   $ sql_datetime_subtype   : chr  NA NA NA NA ...
 ┃   $ char_octet_length      : num  0 0 0 0 255
 ┃   $ ordinal_position       : num  1 2 3 4 5
 ┃   $ nullable               : num  1 1 1 1 1

This is a bit of a function of the silly way the data_type column is formulated in the query ( always NULL ) - we can probably do something better to make sure that the column comes back as numeric, but there is also probably some value in keeping the query as close as the original SQLColumns implementation for the OEM oracle driver.