Re: [PATCH] debuginfod: Support queries for ELF/DWARF sections.

[Aaron Merey <amerey@redhat.com>]

Hi Frank,

On Wed, Sep 28, 2022 at 10:28 AM Frank Ch. Eigler <fche@redhat.com> wrote:
> On Tue, Sep 27, 2022 at 10:10:52PM -0400, Aaron Merey via Elfutils-devel wrote:
>
> > [...]  In order to distinguish between debuginfo and executable
> > files with the same build-id, this function includes a bool
> > parameter use_debuginfo.  If true, attempt to retrieve the section
> > from the debuginfo file with the given build-id. If false, use the
> > executable instead.  [...]
>
> How would a client know which one to use?  Does it provide power or
> benefit to force them to choose (or iterate?).  Is there a scenario
> where the content could be different between the two (if both
> existed)?
>
> If that decisionmaking is not warranted to put upon the shoulders of
> the client, the server could just be asked for a section name "as if
> from an unstripped executable", and let it find that in the executable
> or debuginfo, whereever.

Good point, the server/client should figure this out internally.  On
IRC we also discussed the possible usefulness of client-side
emulation of section queries in case a server isn't built with
_find_section support.  Will update the patch to include these details.

> > [...] Although this patch does not implement it, we could generate
> > .gdb_index on-the-fly if the target file does not contain it.
> > However for large debuginfo files generating the index can take a
> > non-trivial amount of time (ex. about 25 seconds for a 2.5GB
> > debuginfo file).  [...]
>
> Even that is not too bad, considering that the alternative would be
> having to download that 2.5GB file.  I recall you saying that on some
> distros, gdb-index sections are always there anyway, so we wouldn't
> have to rush to implement this feature.

I did a quick experiment checking the debuginfo for the libraries used
by gdb, firefox and qemu-kvm on F36. Out of the 265 files I checked
only 1 (libicudata.so.69.1 debuginfo) didn't contain a .gdb_index
because it strangely does not contain any .debug_* sections at all.

Aaron