Error when non-unique endpoints are passed to period.apply

[joshuaulrich]

A subsetting error will be thrown if there are duplicate entries in the INDEX argument to period.apply(). This is possible if users create custom end points without using the endpoints() function.

x <- .xts(1:10, 1:10)
ep <- c(2, 4, 6, 8, 10)
period.apply(x, c(0, ep, nrow(x)), mean)
# Error in `[.xts`(x, (INDEX[y] + 1):INDEX[y + 1]) : 
#   subscript out of bounds

Possibly worse, you get incorrect results if INDEX is not sorted.

set.seed(21)
EP <- sample(head(ep, -1))
merge(unsorted = period.apply(x, c(0, EP, nrow(x)), mean),
      sorted = period.apply(x, c(0, sort(EP), nrow(x)), mean))
#                     unsorted sorted
# 1969-12-31 18:00:02      4.5    1.5
# 1969-12-31 18:00:04      5.5    3.5
# 1969-12-31 18:00:06      3.5    5.5
# 1969-12-31 18:00:08      5.5    7.5
# 1969-12-31 18:00:10      8.5    9.5

The simplest solution is probably to call xts::isOrdered(INDEX) before looping over it. This solves both problems because isOrdered() returns FALSE if x has duplicate entries. If isOrdered() returns FALSE, we should throw a warning and then call sort and unique on INDEX before proceeding.

We could also prevent any performance hit during internal calls to period.apply() (e.g. apply.daily) by checking for a specific argument passed via ... (e.g. _internal = TRUE). Using a non-syntactic name would prevent accidental matching with arguments intended for FUN.

[joshuaulrich]

The two commits fix the issue, but I'm going to leave this open until I can do some performance benchmarking.

[joshuaulrich]

Closing without adding benchmarks. Correctness first, then performance.