patperry / old-r-frame Goto Github PK

keys(x[y]) should be keys(y) when y is a dataset of keys

replace NA with ""

allow datetime keys: convert Dates to POSIXct (tz = "UTC"); convert POSIXlt to POSIXct, leaving timezone as-is

Currently, this clashes with data.table:

• rowid

These clash with dplyr:

• distinct
• order_by
• slice

Should probably get rid of distinct and add an S3 instance for unique.

May not be worth it for the rest; dplyr and data.table also clash with each other. Users probably won't use two of these packages at the same time anyway.

This should return a vector:

x[cbind(i,j)]

i gives row index, j gives column index

otherwise restrict is not well-defined.

NULL key names should be allowed, otherwise cbind is messy

x[i] where i is logical having same dimensions as x should work

as.matrix.data.frame is broken, need to write our own (https://stat.ethz.ch/pipermail/r-devel/2017-December/075187.html )

once API stabilizes

e.g., if key is int, query is .Machine$integer.max + 1, report not found; can't rely on as.integer because NA might be a key

also applies to character key with invalid UTF-8 or bytes query

make sure argument keys match