Implement ${board}_{exec,spawn} for native-extended-gdbserver

  Hi,

Mark Wielaard is kindly running two Debian buildslaves in our
buildbot, and he noticed that, when testing things under the
native-extended-gdbserver board, there is a huge slowndown due to
several timeouts that are happening.

If you look at the following log:

  <http://gdb-build.sergiodj.net/builders/Debian-i686-native-extended-gdbserver/builds/48/steps/test%20gdb/logs/stdio>

You will notice that this happens because dejagnu is trying to open a
rsh connection to test native-extended-gdbserver, which of course is
wrong, and we end up not testing things properly.

After some investigation (motivated by an idea from Pedro), I found
that our native-extended-gdbserver board does not set the "isremote"
flag properly.  I fixed that by doing two things:

- Defining the ${board}_spawn procedure inside the board file, and

- Moving the ${board}_exec procedure (present at the native-gdbserver
  and native-stdio-gdbserver boards) to the gdbserver-base file.

After that, I am not seeing this failure anymore.

Is this OK to check in?

gdb/testsuite/ChangeLog:
2015-02-03  Sergio Durigan Junior  <sergiodj@redhat.com>

	* boards/gdbserver-base.exp (${board}_exec): New procedure, moved
	from native-gdbserver.exp and native-stdio-gdbserver.exp.
	* boards/native-extended-gdbserver.exp: Do not set isremote using
	board_info.
	(${board}_spawn): New procedure.
	* boards/native-gdbserver.exp (${board}_exec): Moved to
	gdbserver-base.exp.
	* boards/native-stdio-gdbserver.exp (${board}_exec): Likewise.
---
 gdb/testsuite/boards/gdbserver-base.exp            | 12 ++++++++++++
 gdb/testsuite/boards/native-extended-gdbserver.exp | 18 ++++++++++++------
 gdb/testsuite/boards/native-gdbserver.exp          | 12 ------------
 gdb/testsuite/boards/native-stdio-gdbserver.exp    | 12 ------------
 4 files changed, 24 insertions(+), 30 deletions(-)
  

On 02/04/2015 12:20 AM, Sergio Durigan Junior wrote:

> diff --git a/gdb/testsuite/boards/native-extended-gdbserver.exp b/gdb/testsuite/boards/native-extended-gdbserver.exp
> index 57503db..0059c4e 100644
> --- a/gdb/testsuite/boards/native-extended-gdbserver.exp
> +++ b/gdb/testsuite/boards/native-extended-gdbserver.exp
> @@ -23,12 +23,6 @@
>  load_generic_config "extended-gdbserver"
>  load_board_description "gdbserver-base"
>  
> -# By default, dejagnu makes the board remote unless the board name
> -# matches localhost.  Force it to be NOT remote.
> -global board
> -global board_info
> -set board_info($board,isremote) 0

I don't think we can/should remove this.  That'll make
"is_remote" checks all over the testsuite return true,
and thus skip tests that weren't skipped before.  E.g.,
solib-display.exp.

I think we need to instead do here what ${board}_spawn
is already doing.  That is, remove any target variant
specifications from $board, like:

 --- a/gdb/testsuite/boards/native-extended-gdbserver.exp
 +++ b/gdb/testsuite/boards/native-extended-gdbserver.exp
 @@ -27,7 +27,9 @@ load_board_description "gdbserver-base"
  # matches localhost.  Force it to be NOT remote.
  global board
  global board_info
 -set board_info($board,isremote) 0
 +# Remove any target variant specifications from the name.
 +set baseboard [lindex [split $board "/"] 0]
 +set board_info($baseboard,isremote) 0

I'm testing that.

Thanks,
Pedro Alves
  

Pedro Alves <palves@redhat.com> writes:
> On 02/04/2015 12:20 AM, Sergio Durigan Junior wrote:
>
>> diff --git a/gdb/testsuite/boards/native-extended-gdbserver.exp b/gdb/testsuite/boards/native-extended-gdbserver.exp
>> index 57503db..0059c4e 100644
>> --- a/gdb/testsuite/boards/native-extended-gdbserver.exp
>> +++ b/gdb/testsuite/boards/native-extended-gdbserver.exp
>> @@ -23,12 +23,6 @@
>>  load_generic_config "extended-gdbserver"
>>  load_board_description "gdbserver-base"
>>  
>> -# By default, dejagnu makes the board remote unless the board name
>> -# matches localhost.  Force it to be NOT remote.
>> -global board
>> -global board_info
>> -set board_info($board,isremote) 0
>
> I don't think we can/should remove this.  That'll make
> "is_remote" checks all over the testsuite return true,
> and thus skip tests that weren't skipped before.  E.g.,
> solib-display.exp.
>
> I think we need to instead do here what ${board}_spawn
> is already doing.  That is, remove any target variant
> specifications from $board, like:
>
>  --- a/gdb/testsuite/boards/native-extended-gdbserver.exp
>  +++ b/gdb/testsuite/boards/native-extended-gdbserver.exp
>  @@ -27,7 +27,9 @@ load_board_description "gdbserver-base"
>   # matches localhost.  Force it to be NOT remote.
>   global board
>   global board_info
>  -set board_info($board,isremote) 0
>  +# Remove any target variant specifications from the name.
>  +set baseboard [lindex [split $board "/"] 0]
>  +set board_info($baseboard,isremote) 0
>
> I'm testing that.

Eewwww....

[not your patch, just the fact that $board ends up with entire string,
including, e.g., /-m32]

We should document this somewhere, as I can imagine it being
all too easy to continue to just use $board without realizing this.
A little grepping makes me wonder.
E.g., all the procs with ${board} as a prefix.
bash$ grep -F '${board}' */*.exp
proc ${board}_file { dest op args } {
...

and sure enough, those procs get "/-m32" in the name too.
I added
verbose -log "[info procs *${board}*]" to gdbserver-base.exp
and got
native-gdbserver/-m32_file native-gdbserver/-m32_download native-gdbserver/-m32_upload

Yikes.

btw, The name "baseboard" is a bit confusing because I immediately
think of /usr/share/dejagnu/baseboards, which is a different thing.