William Dunlap’s posts on R-help are the place to look for all things related to run lengths. His f7 from this post is
f7 <- function(x){ tmp<-cumsum(x);tmp-cummax((!x)*tmp)}
and in the current situation f7(!x)
. In terms of performance there is
> x <- sample(0:1, 1000000, TRUE)
> system.time(res7 <- f7(!x))
user system elapsed
0.076 0.000 0.077
> system.time(res0 <- cumul_zeros(x))
user system elapsed
0.345 0.003 0.349
> identical(res7, res0)
[1] TRUE