"not in" operator %!in%

[jangorecki]

FR to propose a not-in operator

`%!in%` = function(x, y) !(x%in%y)

Motivation is to have a single call that we could optimise. Otherwise we have to investigate unevaluated expression tree more deeply (and in a much more complex way) to know when user is doing a not-in operation.
Pseudo code in R C api

CAR(x %!in% 1:10)
#`%!in%`
CAR(! x %in% 1:10)
#`!`

[jangorecki]

Actually we still need to handle ! because !x%in% was just one example, another example is !x%like%y (where we would need x%!like%y). It is definitely not robust to ask users to rewrite their code from base R style to our functions so we can optimize them. And it turns out it was not that difficult to handle !. I will leave this issue open because there are upvotes, so it seems such helper function might be useful anyway.

[jaapwalhout]

Such a function is certainly useful! I would prefer to write it as %nin% though, which is easier to type (imho)

[legendre6891]

I would really welcome the addition of %notin%, even if unoptimized; it is quite handy to have.

[mczek]

I've been working on this and I'm wondering if someone can explain something for me. How does this function call get evaluated?
chmatch("a", c(), 5)
It returns 5, but as far as I can tell it dispatches to chmatchMain and it should return a vector allocVector(chin?LGLSXP:INTSXP, xlen) of length 0 which is logical(0) but somehow it returns 5 from R. There's no reference to the nomatch parameter before it returns.

[MichaelChirico]

That's happening here:

data.table/src/chmatch.c

Lines 30 to 36 in 788c585

	const int tablelen = length(table);
	if (tablelen==0) {
	const int val=(chin?0:nomatch), n=xlen;
	for (int i=0; i<n; ++i) ansd[i]=val;
	UNPROTECT(nprotect);
	return ans;
	}

Note that this does match the corresponding match() behavior:

match("a", NULL, 5)
# [1] 5