Re: [PATCH 1/2] gdb: Convert completion tracker to use std types

["Christian Biesinger via gdb-patches" <gdb-patches at sourceware dot org>]

On Fri, Jan 24, 2020, 20:14 Tom Tromey <tom@tromey.com> wrote:

> >>>>> "Andrew" == Andrew Burgess <andrew.burgess@embecosm.com> writes:
>
> Andrew> For now then I'm holding the raw 'char *' within the unordered_map,
> Andrew> and rely on the completion_tracker object to delete the data if
> Andrew> needed, or to ensure that the data is passed over to readline,
> which
> Andrew> will do the deletion for us.
>
> Seems reasonable enough.
>
> Andrew> +  for (const auto &p : m_entries_hash)
> Andrew> +    {
> Andrew> +      xfree ((char *) p.first);
> Andrew> +      xfree ((char *) p.second);
>
> Maybe const_cast would be better?
>
> Andrew> +  /* A map of completions.  The key is the completion, and the
> value is the
> Andrew> +     string to be used to compute the lowest common denominator.
> Both the
> Andrew> +     key and the value are owned by the completion_tracker class
> while
> Andrew> +     being held in this map, as such completion_tracker must free
> these
> Andrew> +     strings when finished with them, or pass ownership to
> someone else who
> Andrew> +     will free them.  */
> Andrew> +  typedef std::unordered_map<const char *, const char *,
> std::hash<std::string>,
> Andrew> +                            std::equal_to<std::string>>
> completion_set;
>
> Does using std::string here mean that a temporary std::string will be
> made for each insertion in the map?  And also for comparisons?
> That seems expensive if so.
>

Perhaps that's finally a use case to add hash<gdb::string_view>!


> Tom
>