This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFA take 6] Allow setting breakpoints on inline functions (PR 10738)


> Date: Thu, 15 Mar 2012 18:30:43 +0000
> From: Pedro Alves <palves@redhat.com>
> CC: Gary Benson <gbenson@redhat.com>, dje@google.com,
>         jan.kratochvil@redhat.com, gdb-patches@sourceware.org, mark@klomp.org
> 
> On 03/15/2012 06:14 PM, Eli Zaretskii wrote:
> 
> >> Date: Thu, 15 Mar 2012 18:10:02 +0000
> >> From: Gary Benson <gbenson@redhat.com>
> >> Cc: dje@google.com, jan.kratochvil@redhat.com, gdb-patches@sourceware.org,
> >>         mark@klomp.org
> >>
> >>>> "Do not reject possibly inconsistent .gdb_index sections."
> >>>
> >>> The meaning of that is that the sections being skipped are
> >>> inconsistent within themselves.  If that's really what you meant,
> >>> I'm fine with the change.
> >>
> >> The issue is that with older index section the information in the
> >> .gdb_index sections is not consistent with the information that GDB
> >> would generate from the DWARF.
> > 
> > Why does this happen?  Is the information in those sections
> > inaccurate?
> 
> A more high level alternative explanation to Gary's would be
> something like:
> 
> Older GDB versions didn't use all the info from DWARF that the new
> versions do, and so not all the info needed by newer GDBs is in
> the older index sections (we only put there what we need).  Using the
> index is mutually exclusive with fetching the info out of DWARF.
> So if GDB loads an older index, there are bits of info that are
> missing.  And, with those missing, a newer GDB will present a
> worse debugging experience in terms of features and correctness
> than if it didn't use the index at all, but fetched all it needed
> from the DWARF.
> 
> Hope I got that right.

Thanks.  So I think "incomplete" was a better word.  But if for some
reason we don't want that, how about "inaccurate"?  E.g.

  Do not reject obsolete .gdb_index sections with possibly inaccurate info.

However, if everyone else is tired of bikeshedding, go ahead with
whatever you like.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]