Zap inline objects from backtraces

[lionel-]

Closes #1069.
Closes r-lib/testthat#1223.

Inline objects are replaced by their pillar type sum:

call <- expr(foo(!!(1:2)))
call_zap_inline(call)
#> foo(`<int>`)

[hadley]

I find this slightly easier to understand, but it's not that important

Suggested change

	return(x)
	}
	if (is_symbol(x) || is_syntactic_literal(x)) {
	x
	x
	} else if (is_symbol(x) || is_syntactic_literal(x)) {
	x

[lionel-]

The idea here is to keep the terminating and recursive branches separate.

[hadley]

Could also use do.call() here, since that's the most common way that users will create inlined objects.

[hadley]

Maybe something like along these lines:

e <- environment()
f <- function() do.call("g", list(1:1e6 + 1))
g <- function(...) h()
h <- function() rlang::trace_back(e)
lobstr::obj_size(f())
#> 8,014,888 B

^{Created on 2021-09-14 by the reprex package (v2.0.0)}

(But using something other than obj_size())

[hadley]

Maybe

trace <- f()
trace$calls[[3]]

?

[lionel-]

Good idea, I've added a do.call() and size check in the test.

[hadley]

Again, not very important, but why does this change?

[lionel-]

The injected handler is now a symbol. This allows as_label() to keep the whole call instead of eliding.