assign_in() NULL should assign rather than remove

[lionel-]

The modfied object returned by pluck<- should always have the same length structure.

[lionel-]

Document difference with [[<-: no removal, no expansion.

[lionel-]

After feedback from #634, maybe we should let puck create new elements — at least by name, it feels like assigning by index should be stricter (semantic difference between vector and associative containers).

This means we'd allow removing elements with zap(), but only by name?

What do you think @hadley?

[hadley]

I think it seems like a nice principle that pluck<-(x, i, value) have the same sizes and types as x. Assigning deeply into structures where bits might not exist seems risky to me and I think you'd be better off first normalising your data structure with some other tool.

[lionel-]

In the shallow case it seems reasonable to do assign_in(x, "elt", value) and expect new elements to be added. In that case it's more of a $<- action than a [[<- one.

[hadley]

I think this needs more thought about what our desired $<- vs [[<- vs [<- semantics are

[hadley]

Reprex:

library(purrr)

x <- list(a = 1, b = list(c = 1))
pluck(x, "a") <- NULL
str(x)
#> List of 1
#>  $ b:List of 1
#>   ..$ c: num 1

pluck(x, "b", "c") <- NULL
str(x)
#> List of 1
#>  $ b: Named list()

^{Created on 2022-08-25 by the reprex package (v2.0.1)}

Definitely feels like purrr is standardising on using zap() to delete.