data.table with extended `==` function not working in R-CodePudding

This is a follow-up question below.

How to extend `==` behavior to vectors that include NAs?

In the link, suppose we would like to compare

a = c(1,2,NA)
b = c(1,3,NA)

Using the function of "%==%" <- function(a, b) (!is.na(a) & !is.na(b) & a==b) | (is.na(a) & is.na(b)), and we have the following result

a %==% b
# TRUE FALSE  TRUE

My question is how to use %==% in the data.table::data.table function. Why the following code does not work properly? The first two results are good, but it seems that %==% cannot be used with operations, i.e. the a a2 below.

dt_a = data.table(a = a, a2 = a a)
dt_a[,a %==% b] 
# TRUE FALSE  TRUE
dt_a[,a2 %==% b]
# FALSE FALSE  TRUE
dt_a[,a a2 %==% b]
# 1  2 NA # this line of result is confusing.

CodePudding user response：

It's a problem of operator precedence. You would need

dt_a[,(a a2) %==% b]

The fake equals has a different precedence than a normal == what you write originally was equivalent to

dt_a[,a (a2 %==% b)]

Addition operators are higher precedence than test for equality but lower precedence than custom infix %% functions. See the ?Syntax help page for the full details.

Note this doesn't have anything to do with data.table specifically. Same things happens outside

a <- c(1,2,NA)
b <- c(1,3,NA)
a2 <- a  a
a a2 %==% b
# [1]  1  2 NA