Re: [PATCH 4/5] Add "target:" filename handling to find_separate_debug_file

[Gary Benson <gbenson at redhat dot com>]

Gary Benson wrote:
> Gary Benson wrote:
> > Pedro Alves wrote:
> > > On 06/16/2015 10:42 AM, Gary Benson wrote:
> > > > This commit updates find_separate_debug_file to handle
> > > > filenames prefixed with "target:".  The same-directory and
> > > > DEBUG_SUBDIRECTORY locations are checked with the prefix if
> > > > supplied.  The debugdir location is checked both with and
> > > > without the prefix if one is supplied.  This makes GDB able
> > > > to fetch separate debug files from remote targets and from
> > > > inferiors in containers.
> > > > 
> > > > +      /* Try the same location but without TARGET_SYSROOT_PREFIX
> > > > +	 (i.e. on the local filesystem).  */
> > > > +      debugfile = build_debug_file_name (no_prefix, debugdir, dir,
> > > > +					 debuglink, NULL);
> > > 
> > > Given that we have a CRC to match, shouldn't we try the local
> > > filesystem first, avoiding the (potentially slow) remote
> > > fetching in the case the files on the container/remote are
> > > the same of the host's?  (which I think happens often with
> > > containers).
> > 
> > I think for both remote and container cases the answer to "are
> > they the same" is no more than "they might be".
> > 
> > With containers, "target:" fetching shouldn't be very much slower
> > than local-filesystem fetching, just a couple more operations at
> > file open time.  With remote it will be slower, but less likely
> > to be the same (maybe!)
> > 
> > That was my reasoning.  I can switch it round if you want though.
> > Let me know what you'd like...
> 
> Given the slow remote transfers thread on the other list [1],
> I will switch it round (i.e. try locally first).
> --
> [1] https://sourceware.org/ml/gdb/2015-07/msg00038.html

Actually, I now discover that the reason I did it "target:" first
was that doing it the other way around clutters up the display
with loads of warnings if the two systems are different:

  Reading symbols from target:/usr/lib64/libpython2.6.so.1.0...
  warning: the debug information found in "/usr/lib/debug/usr/lib64/libpython2.6.so.1.0.debug" does not match "target:/usr/lib64/libpython2.6.so.1.0" (CRC mismatch).
  Reading symbols from target:/usr/lib/debug/usr/lib64/libpython2.6.so.1.0.debug...done.
  done.
  Reading symbols from target:/lib64/libpthread.so.0...
  warning: the debug information found in "/usr/lib/debug/lib64/libpthread-2.12.so.debug" does not match "target:/lib64/libpthread.so.0" (CRC mismatch).
  Reading symbols from target:/usr/lib/debug/lib64/libpthread-2.12.so.debug...done.
  done.
  ...etc...

If you're remote debugging over a slow link and have set gdb_sysroot
to some local copy then there will be no "target:" paths here so there
will be no slowdown with a target-first search.

If you're remote debugging with local debuginfo there will be some
slowdown due to failing vFile:open: packets.  One round trip per
executable/library to be precise.  Is this acceptable to you?

Cheers,
Gary

-- 
http://gbenson.net/