Re: [PATCH 1/4] gdb: add intern method to objfile_per_bfd_storage

[Christian Biesinger <cbiesinger@google.com>]

On Tue, Mar 23, 2021 at 4:49 PM Simon Marchi <simon.marchi@polymtl.ca> wrote:
>
> On 2021-03-23 10:47 a.m., Christian Biesinger via Gdb-patches wrote:
> > On Mon, Mar 22, 2021 at 8:34 PM Simon Marchi via Gdb-patches
> > <gdb-patches@sourceware.org> wrote:
> >>
> >> From: Simon Marchi <simon.marchi@polymtl.ca>
> >>
> >> This allows factoring out the internal implementation details that are
> >> currently in the two objfile::intern methods.
> >>
> >> gdb/ChangeLog:
> >>
> >>         * objfiles.h (struct objfile_per_bfd_storage) <intern>: New
> >>         method.
> >>         (struct objfile) <intern>: Use
> >>         objfile::objfile_per_bfd_storage::intern.
> >>
> >> Change-Id: Ifd54026c5efaeffafac9b84ff84c199acc7ce78a
> >> ---
> >>  gdb/objfiles.h | 13 ++++++++++---
> >>  1 file changed, 10 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/gdb/objfiles.h b/gdb/objfiles.h
> >> index 41f8fc913d8..d9aa06636f5 100644
> >> --- a/gdb/objfiles.h
> >> +++ b/gdb/objfiles.h
> >> @@ -270,6 +270,14 @@ struct objfile_per_bfd_storage
> >>
> >>    ~objfile_per_bfd_storage ();
> >>
> >> +  /* Intern STRING in this object's string cache and return the unique copy.
> >> +     The copy has the same lifetime as this object.  */
> >> +
> >> +  const char *intern (gdb::string_view str)
> >> +  {
> >> +    return (const char *) string_cache.insert (str.data (), str.size () + 1);
> >
> > Is this guaranteed to be nullterminated even without calling c_str()?
> >
> > Christian
>
> From what I remember from last time something like this came up (I can't
> remember the context) is that we'd only use string_view for strings that
> are null-terminated, even though string_view doesn't guarantee it.  Is
> that the case?  Or is it a bad use case for string_view?

Hm, I'm not sure we concluded that in the general case. It came up
specifically for general_symbol_info::compute_and_set_names, which can
handle the case where it may or may not be nullterminated depending on
a parameter. IMO you should at least comment that the string view must
be nullterminated here.

> In this particular case, callers of this intern method have a C string
> and an std::string, so we know they are null-terminated.

OK, makes sense.

Christian