Why do new_pillar_shaft() and new_pillar_shaft_simple() default min_width to width?

[DavisVaughan]

In bench, there is a custom expression class called bench_expr that displays the captured expressions

We have a pillar_shaft() method for this that looks like:

pillar_shaft.bench_expr <- function(x, ...) {
  pillar::new_pillar_shaft_simple(format.bench_expr(x), align = "left", ...)
}

It seems like the way we defined this method is actually the cause of this issue
r-lib/bench#94

i.e. if you have a very long bench expression then it won't get truncated like a normal character column would. It seems like this is because min_width defaults to width, disallowing any truncation.

It seems like I can fix it by doing this:

pillar_shaft.bench_expr <- function(x, ...) {
  pillar::pillar_shaft(format.bench_expr(x), ...)
}

i.e. convert to character and then use the standard pillar_shaft.character() method.

The pillar_shaft.character() method respects pillar.min_chars through get_pillar_option_min_chars() so the min_width ends up being a good default of 3 by doing this.

I wonder if the default of new_pillar_shaft() and new_pillar_shaft_simple() should also respect this option rather than defaulting to width?

[DavisVaughan]

Maybe width is a good default for most types (because, for example, you want to see the full number with a double column), and character vectors are just a bit special?

In that case, my solution above is pretty similar to what is done for factors, so I feel pretty good about it

> pillar:::pillar_shaft.factor
function (x, ...) 
{
    pillar_shaft(as.character(x), ...)
}

Feel free to close if you agree with my thoughts here

[krlmlr]

Thanks. The solution applied in bench looks good to me. With rlang::expr_deparse(width = ...), you could perhaps truncate in a nicer way (if not done already).