R的清洁，简单的功能工厂

Question

Short example. I am exploring the behavior of a function by testing it with different "specs", f(spec) . I wrote down one spec by hand, spec1 , and am creating new specs as variations on it. To do this, I decided to write a function:

spec1 = list(fy = list(a = 1), fx = list(f1 = function(x) 10-x, f2 = function(x) 2-x))

make_spec = function(f = function(x) 10-x, xtheta = 2)
    list(fy = list(a = 1), fx = list(f1 = f, f2 = function(x) xtheta-x))

res1 = make_spec()

# first problem: they don't match

    all.equal(res1,spec1)
    # [1] "Component “fx”: Component “f2”: target, current do not match when deparsed"
    # ^ this happens, even though...
    res1$fx$f2(4) == spec1$fx$f2(4)
    # TRUE

# second problem: res1 is fugly

    res1
    # $fy
    # $fy$a
    # [1] 1
    # 
    # 
    # $fx
    # $fx$f1
    # function (x) 
    # 10 - x
    # <environment: 0x000000000f8f2e20>
    # 
    # $fx$f2
    # function (x) 
    # xtheta - x
    # <environment: 0x000000000f8f2e20>

    str(res1)
    # even worse

My goals for make_spec are...

1. all.equal(spec1, res1) and/or identical(spec1, res1)
2. for str(res1) to be human-readable (no <environment: ptr> tags or srcfilecopy )
3. to avoid substitute and eval altogether if possible (not a high priority)
4. to avoid writing out the second arg of substitute (see "full" example below)

Is there an idiomatic way to achieve some or all of these goals?

---

Full example. I'm not sure if the example above fully covers my use case, so here's the latter:

spec0 = list(
    v_dist = list(
        pdf  = function(x) 1,
        cdf  = function(x) x,
        q    = function(x) x,
        supp = c(0,1)
    )
    ,
    ucondv_dist = {
        ucondv_dist = list()
        ucondv_dist$condmean    = function(v) 10-v
        ucondv_dist$pdf         = function(u,v) dnorm(u, ucondv_dist$condmean(v), 50)
        ucondv_dist$cdf         = function(u,v) pnorm(u, ucondv_dist$condmean(v), 50)
        ucondv_dist
    }
)

make_spec = function(ycondx_condmean = function(x) 10-x, ycondx_sd = 50){

  s = substitute(list(
    x_dist = list(
      pdf  = function(x) 1,
      cdf  = function(x) x,
      q  = function(x) x,
      supp = c(0,1)
    )
    ,
    ycondx_dist = {
      ycondx_dist = list()
      ycondx_dist$condmean  = ycondx_condmean
      ycondx_dist$pdf     = function(u,v) dnorm(u, ycondx_dist$condmean(v), ycondx_sd)
      ycondx_dist$cdf     = function(u,v) pnorm(u, ycondx_dist$condmean(v), ycondx_sd)
      ycondx_dist
    }
  )
  , list(ycondx_condmean=ycondx_condmean, ycondx_sd = ycondx_sd))

  eval(s, .GlobalEnv)
}

res0 = make_spec()

---

Side note . I don't know if "function factory" is the right term here, since I am not a computer scientist, but it seems related. I found only a paragraph on the concept related to R .

Answer 1

The enclosing environments of the functions are different leading to the difference in output/difference in deparsing. So, there are two things to do to get the desired output:

• make the environments the same
• substitute the variables from the enclosing environments into the function bodies.

However, doing it this way you get a double dose of the eval/substitute you didn't want, so maybe there would be an alternative.

make_spec <- function(f = function(x) 10-x, xtheta = 2) {
  e <- parent.frame()
  fixClosure <- function(func)
    eval(eval(substitute(substitute(func)), parent.frame()), e)

  list(fy = list(a = 1), fx = list(
    f1 = fixClosure(f), 
    f2 = fixClosure(function(x) xtheta-x)
  ))
}

spec1 <- list(fy = list(a = 1), fx = list(f1 = function(x) 10-x, f2 = function(x) 2-x))
res1 <- make_spec()

all.equal(res1, spec1)
[1] TRUE