Include value in error message about invalid value from register_method()

[t-kalinowski]

Closes #447

[hadley]

Hmmmm, I wonder if we should add a error wrapper in method<- since it has a bit more context. I think ideally I'd want

foo <- function() {}
method(foo, class_character) <- function(...) {}
#> Error: Failed to register <character> method for <foo> generic
#> `generic` is a function, but not an S3 generic function

But looking at the traceback, maybe we can't get the name of the generic because of the *tmp* substitution?

[t-kalinowski]

I don't think it's possible to get the original expression for x in a call like foo(x) <- through introspection.

> `foo<-` <- function(x, value) {
+   print(sys.calls())
+ }
> 
> x <- function() {}
> 
> foo(x) <- "some val"
[[1]]
`foo<-`(`*tmp*`, value = "some val")

[hadley]

Yeah, so the best we could do is mention the method. Maybe something like this?

foo <- function() {}
method(foo, class_character) <- function(...) {}
#> Error: Failed to register <character> method for generic
#> `generic`, function(...) is a function, but not a generic

[hadley]

I think displaying the arguments, without the body, would be sufficient. (Like args(mean) but without the NULL).

[t-kalinowski]

function (x, ...) doesn't seem like a very useful breadcrumb, especially if we can easily include the head of the first expression in the function body.

[hadley]

I guess I'm worried about this spitting out a very large function body, making it hard to see the actual error message.

[t-kalinowski]

width.cutoff = 100L

maybe we truncate this more/better?

[hadley]

I think that just determines the line length?

[t-kalinowski]

TIL! (I had thought width.cutoff capped nchar() of the returned string)

Maybe something like:

deparse_trunc <- function(x, width, collapse = "\n") {
  x <- deparse1(x, collapse)
  if (nchar(x)> width) 
     x <- sprintf("%s....", substr(x, 0, width-4))
  x
}

which gives:

> cat(deparse_trunc(plot.default, 80))
function (x, y = NULL, type = "p", xlim = NULL, ylim = NULL, log = "", main ....

[hadley]

Yeah, that sounds good to me.