[Bug] LOCF roll gives incorrect values when the value of the key is of very large order

[sainathadapa]

library(data.table)
#data.table 1.9.7  For help type ?data.table or https://github.com/Rdatatable/data.table/wiki
#The fastest way to learn (by data.table authors): https://www.datacamp.com/courses/data-analysis-the-data-table-way
options(digits=16)

seed <- data.table(x = c(0.5, 1, 1.008, 1.011), y = c('a', 'b', 'c', 'd'), key='x')
data <- data.table(x = c(0.5, 1.011), z = c('e', 'f'), key='x')
data[seed, mult = "first", roll = T]
#       x z y
#1: 0.500 e a
#2: 1.000 e b
#3: 1.008 e c
#4: 1.011 f d
seed <- data.table(x = c(0.5, 1, 1.008, 1.011) + 144557957, y = c('a', 'b', 'c', 'd'), key='x')
data <- data.table(x = c(0.5, 1.011) + 144557957, z = c('e', 'f'), key='x')
data[seed, mult = "first", roll = T]
#               x z y
#1: 144557957.500 e a
#2: 144557958.000 e b
#3: 144557958.008 e c
#4: 144557958.011 f d
seed <- data.table(x = c(0.5, 1, 1.008, 1.011) + 1445579573, y = c('a', 'b', 'c', 'd'), key='x')
data <- data.table(x = c(0.5, 1.011) + 1445579573, z = c('e', 'f'), key='x')
data[seed, mult = "first", roll = T]
#                x z y
#1: 1445579573.500 e a
#2: 1445579574.000 e b
#3: 1445579574.008 f c
#4: 1445579574.011 f d

Check the 3rd and 4th rows in the last case against the answers from the 1st and 2nd cases. 'f' value in z column should occur only in the last row for all the cases.

sessionInfo()
#R version 3.2.2 (2015-08-14)
#Platform: x86_64-pc-linux-gnu (64-bit)
#Running under: Ubuntu 14.04.3 LTS
#
#locale:
 #[1] LC_CTYPE=en_IN.UTF-8       LC_NUMERIC=C               LC_TIME=en_IN.UTF-8        #LC_COLLATE=en_IN.UTF-8     LC_MONETARY=en_IN.UTF-8    LC_MESSAGES=en_IN.UTF-8   
 #[7] LC_PAPER=en_IN.UTF-8       LC_NAME=C                  LC_ADDRESS=C               #LC_TELEPHONE=C             LC_MEASUREMENT=en_IN.UTF-8 LC_IDENTIFICATION=C       
#
#attached base packages:
#[1] stats     graphics  grDevices utils     datasets  methods   base     
#
#other attached packages:
#[1] data.table_1.9.7
#
#loaded via a namespace (and not attached):
#[1] tools_3.2.2  chron_2.3-47

Edit: Tested on the latest development version.

[arunsrinivasan]

Do setNumericRounding(0L) so that the groups aren't computed on the (implicitly) rounded column. See ?setNumericRounding.

I'll close this once there's a nicer way of handling this / better way of keeping people informed (thro' vignettes).

[sainathadapa]

Thanks. I didn't know about that option before today.

[arunsrinivasan]

Updated #944 to come back to this.