splicemarks an object to be spliced. It is equivalent to using
!!!in a function with tidy dots semantics.
dots_splice(..., .ignore_empty = c("trailing", "none", "all"))
A list to splice.
Arguments with explicit (
dots_list()) or list (
dots_splice()) splicing semantics. The contents of spliced arguments are embedded in the returned list.
Whether to ignore empty arguments. Can be one of
"trailing", only the last argument is ignored if it is empty.
dots_list()but automatically splices list inputs.
Standard splicing versus quoting splicing
In quoting functions
!!!disaggregates its argument (let's call it
x) into as many objects as there are elements in
quo(foo(!!! c(1, 2)))is completely equivalent to
quo(foo(1, 2)). The creation of those separate objects has an overhead but is typically not important when manipulating calls because function calls typically take a small number of arguments.
In standard functions, disaggregating the spliced collection would have a negative performance impact in cases where
dots_list()is used to build up data structures from user inputs. To avoid this spliced inputs are marked with
splice()and the final list is built with (the equivalent of)
Most of the time you should not care about the difference. However
if you use a standard function taking tidy dots within a quoting
!!! operator will disaggregate its argument because
the behaviour of the quasiquoting function has priority. You might
then observe some performance cost in edge cases. Here is one
example where this would happen:
purrr::rerun(10, dplyr::bind_rows(!!! x))
purrr::rerun() is a quoting function and
a standard function. Because
bind_rows() is called inside
rerun(), the list
x will be disaggregated into a pairlist of
arguments. To avoid this you can use
dots_splice()is in questioning stage. It is part of our experiments with dots semantics. Compared to
dots_splice()automatically splices lists. We now lean towards adopting a single type of dots semantics (those of
dots_list()) where splicing is explicit.
splice()is in questioning stage. It is not clear whether it is really needed as there are other ways to avoid the performance issue discussed in the section above.