Re: [patchv3 8/8] Tests for validate symbol file using build-id

[Pedro Alves <palves at redhat dot com>]

On 02/28/2014 08:33 PM, Jan Kratochvil wrote:
> On Fri, 28 Feb 2014 14:28:36 +0100, Pedro Alves wrote:
>> On 02/27/2014 09:33 PM, Jan Kratochvil wrote:
>>> +
>>> +file delete -force -- "${binlibfiledirrun}"
>>> +file mkdir "${binlibfiledirrun}"
>>
>> Looks like as is, the test should be skipped when remote host testing?
> 
> The file is already skipped for non-gdbserver (=NAT) mode as linux-nat does
> not support build-id reporting...

That's actually orthogonal to what I was pointing at.  With remote host
testing, you can have:

 GNU/Linux build x Windows host x Linux GDBserver target

Use of tcl's "file" procedures obviously break that scenario, as those
run on "build".  The fix is usually to use "remote_file host" and
'remote_exec host "mkdir"' etc. instead.  (See e.g., Doug's recent
change to argv0-symlink.exp.)

> 
> I have added there:
> 	if { [board_info target sockethost] != "localhost:" } {
> 	  # The testcase below could be fixed for remote targets.
> 	  untested "only gdbserver on localhost is supported (found [board_info target sockethost])"
> 	  return -1
> 	}

Not sure.  I think I'd rather leave this out, so that someone else
doing remote target testing might perhaps get motivated to address it.
Might it be just a gdb_load_shlibs call is missing?  Can't easily tell
from just a skim over the test.

As is, I think the test is being skipped with
--target_board=native-extended-remote.exp, due to
the !is_remote check.  Sounds like gdb_is_target_remote
from qtro.exp would be a better fit, or IOW, time to move
that to gdb.exp ?

> Someone on gdb-patches was fixing testcases from
> time to time to run on real non-localhost remote configurations.  When such
> configuration is already setup it is easier to fix all such skipped testcases.

FWIW, we have "--host_board=local-remote-host" nowadays so anyone
can easily try, but I don't know how well that works.

-- 
Pedro Alves