Re: [PATCH v2 3/5] Extended-remote support for exec event tests

[Pedro Alves <palves at redhat dot com>]

On 07/31/2015 12:19 AM, Don Breazeal wrote:

> This patch updates several exec-related tests and some of the 
> library functions in order to get them running with extended-remote.
> There were three changes that were required, as follows:
> 
> In gdb.base/foll-exec.exp, the proc 'zap_session' is used repeatedly
> to reset the state of the debugger before the next test.  Part of
> that procedure is to 'set exec-file'.  For remote targets, it is
> necessary to also 'set remote exec-file' to achieve the same
> effect (and execute the correct binary file in the subsequent test).

This assumes local gdbserver testing.  This is something that
should be done in the target board.  Look for exec-file in
gdb/testsuite/boards/native-extended-gdbserver.exp.
Maybe we should just clean_restart instead ?

> 
> In gdb.base/pie-execl.exp, there is an expect statement with an
> expression that is used to match output from both gdb and the
> program under debug.  For the remote target, this had to be 
> split into two expressions, using $inferior_spawn_id to match
> the output from the program.
> 
> Because I had encountered problems with extended-remote exec events
> in non-stop mode in my manual testing, I added non-stop testing to
> the non-ldr-exc-[1234].exp tests.  In order to set non-stop mode
> for remote targets, it is necessary to 'set non-stop on' after gdb
> has started, but before it connects to gdbserver.  The non-ldr-...
> tests call 'clean_restart' in between tests, and it eventually calls
> 'gdb_start' which starts gdb and gdbserver and connects them.  By
> adding a stop mode argument to clean_restart and gdb_start (in
> both lib/gdb.exp and boards/native-extended-gdbserver.exp), it was
> possible to set non-stop mode for remote targets.  Since the
> arguments have a default value "all-stop", and only have an effect
> when "non-stop" is passed, these changes do not affect any existing
> test behavior.

I don't think we can do this, because gdb_start is an overridable
function (see comment above it).  You can instead append
set non-stop on to GDBFLAGS.  E.g.,:

save_vars { GDBFLAGS } {
    append GDBFLAGS " -ex \"set non-stop $nonstop\""
    clean_restart
}

> diff --git a/gdb/testsuite/gdb.threads/non-ldr-exc-1.exp b/gdb/testsuite/gdb.threads/non-ldr-exc-1.exp
> index 69e5cc6..147e7f3 100644
> --- a/gdb/testsuite/gdb.threads/non-ldr-exc-1.exp
> +++ b/gdb/testsuite/gdb.threads/non-ldr-exc-1.exp
> @@ -28,11 +28,11 @@ if {[gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" "${binfile}" executab
>      return -1
>  }
>  
> -proc do_test { lock_sched } {
> -    with_test_prefix "lock-sched$lock_sched" {
> +proc do_test { lock_sched stop_mode } {
> +    with_test_prefix "lock-sched$lock_sched,$stop_mode" {
>  	global executable
>  
> -	clean_restart ${executable}
> +	clean_restart ${executable} $stop_mode
>  
>  	if ![runto_main] {
>  	    return -1
> @@ -48,11 +48,17 @@ proc do_test { lock_sched } {
>  	    gdb_test_no_output "set scheduler-locking on"
>  	}
>  
> +	if { $stop_mode == "non-stop" } {
> +	    gdb_test "thread 2" "Switching.*"
> +	}
> +
>  	gdb_test "continue" \
>  	    ".*is executing new program.*Breakpoint 1, main.* at .*" \
>  	    "continue over exec"
>      }
>  }
>  
> -do_test 0
> -do_test 1
> +do_test 0 "all-stop"
> +do_test 1 "all-stop"
> +do_test 0 "non-stop"
> +do_test 1 "non-stop"

Please use foreach.  E.g.,

foreach nonstop {"on" "off"} {
  foreach schedlock {"on" "off"} {
     do_test ...
  }
}

Note that schedlock on has no effect in non-stop mode.
Maybe if !lock_sched && nonstop, we could issue "continue -a"
instead of continue.

> diff --git a/gdb/testsuite/gdb.threads/non-ldr-exc-2.exp b/gdb/testsuite/gdb.threads/non-ldr-exc-2.exp
> index 9386153..748ff11 100644
> --- a/gdb/testsuite/gdb.threads/non-ldr-exc-2.exp
> +++ b/gdb/testsuite/gdb.threads/non-ldr-exc-2.exp
> @@ -29,18 +29,26 @@ if {[gdb_compile_pthreads "${srcdir}/${subdir}/${srcfile}" "${binfile}" executab
>      return -1
>  }
>  
> -proc do_test { lock_sched } {
> -    with_test_prefix "lock-sched$lock_sched" {
> +proc do_test { lock_sched stop_mode } {
> +    with_test_prefix "lock-sched$lock_sched,$stop_mode" {
>  	global executable
>  
> -	clean_restart ${executable}
> +	clean_restart ${executable} $stop_mode
>  
>  	if ![runto_main] {
>  	    return -1
>  	}
>  
>  	gdb_breakpoint [gdb_get_line_number "break-here"]
> -	gdb_continue_to_breakpoint "break-here" ".* break-here .*"
> +	gdb_test_multiple "continue" "continue to breakpoint" {
> +	    -re ".*Breakpoint.*break-here.*" {

This doesn't expect the prompt, which is going to be
racy -- the following gdb_test may fail if this test manages
to leave the prompt in the expect buffer.  What motivated this change?

Thanks,
Pedro Alves