I think instead of a std::vector<string_file>, what I meant was just use
> a single string_file.
>
> The caller can make it, using something like:
>
>   string_file file (gdb_stderr->can_emit_style_escape ());
>
> Alexandra> +      string_file warnings (true);
>
> ... since I think 'true' isn't always correct.
>

When is it not true?

>
> Alexandra> +      warnings.printf ("\"%ps\":separate debug info file has
> no debug info",
> Alexandra> +                       styled_string (file_name_style.style (),
> Alexandra> +                                      bfd_get_filename
> (abfd.get ())));
>
> Then this printf can either add a 'warning:' prefix like warning() does,
> or we can add an overload of "warning" that takes a ui_file as its first
> parameter.
>

So you would prefer me to use a single string_file instead of a vector of
those. That means I'll have just a single string and I'll be appending each
new warning to it? I used a vector because I don't know how to easily split
the warnings when printing them. Do you mean I should add an overload of a
"warning" that would be capable of splitting strings? But why is such a
solution better then simply using a vector of string_file?

Thank you for the review,
Alexandra