new.pim.formula completely ignores the left hand side!

[JoFAM]

myPO <- function(x){
  lx <- L(x)
  rx <- R(x)
  n <- length(lx)

  out <- rep(-2, n)
  out[lx > rx] <- 2
  return(out)
}

xx <- rnorm(20)
yy <- xx + runif(20,-0.1,0.1)

pe <- new.pim.env(environment(), compare = "unique",
                  vars = c("xx","yy"))
new.pim.formula(myPO(yy) ~ L(xx) - R(xx), pe)

returns :

PIM formula without intercept: myPO(yy) ~ L(xx) - R(xx) <environment: 0x0000000016c09a38>

Left hand side: PO(L(yy), R(yy))

[JoFAM]

Tracked this down to new.pim.formula not recognizing myPO as a special and hence deciding that has.funs.lhs <- FALSE in new.pim.formula.

Solving this will need a mechanism to register new specials, and tie them to the correct environment so they can find the correct L() and R() functions.

Terms object can have an attribute "specials" that is also used for splines. Might pay off to take a look at that one in order to get pim in a better shape.