Re: [PATCH 4/4] testsuite: Add option to capture gdbserver debug

[Alan Hayward <Alan dot Hayward at arm dot com>]

(Patches 1 to 3 pushed).

> On 16 Apr 2019, at 20:49, Tom Tromey <tom@tromey.com> wrote:
> 
>>>>>> "Alan" == Alan Hayward <Alan.Hayward@arm.com> writes:
> 
> Alan> Add board option which enables gdbserver debug and sends it to the
> Alan> file gdbserver.log, located in the output directory for the current
> Alan> test.  Document this.
> 
> Alan> Add debug versions of the native gdbserver board files.
> 
> Alan> Disable tspeed.exp when debugging to prevent the log file filling
> Alan> many gigabytes then timing out.
> 
> Thanks.
> 
> Alan> +gdbserver,debug
> Alan> +
> Alan> +  When set gdbserver debug is outputed to the file gdbserver.log in the test
> 
> I think it should say "is sent" rather than "is outputed”.

Done.

> Alan> diff --git a/gdb/testsuite/boards/native-extended-gdbserver-debug.exp b/gdb/testsuite/boards/native-extended-gdbserver-debug.exp
> 
> I wonder if a new board is needed for this?
> Could it be done some other way, like a command-line setting?

Agreed, it’s a little awkward. I’ve been playing round with an alternative
implementation and have added it an environment variable (in addition to the
board setting).

So, now you can do:
	make check GDBSERVER_DEBUG=all

Which is much nicer. However, see final comment.

> 
> TBH I'm not sure what the typical approach is for something like this.
> 
> Alan>  	if {![info exists gdbserver_reconnect_p] || !$gdbserver_reconnect_p} {
> Alan>  	    # GDB client could accidentally connect to a stale server.
> Alan> -	    # append gdbserver_command " --debug --once"
> Alan>  	    append gdbserver_command " --once"
> 
> Was this intentional?

Yes, I removed this commented out line. I suspect it was there so that you could
quickly switch over to using "—debug". With my changes you shouldn’t need it.

> 
> 
> One random thought I had about this series is that it would be nice to
> have a way to get "set remotelogfile" output in the test directory.
> That way a failing test could be re-run with gdbreplay without much
> trouble.

Yes, I think it’d be possible using a similar approach.  You’d end up with another
environment variable, REPLAY_LOG=1 or something.  My only concern is that it’s
starting down the road of adding more environment vars to the make line - which
people might not want?


I’ll post the updated version as a V2 so that I can use git send-email.


Alan.