Re: [committed][gdb/testsuite] Split up multi-exec test-cases

[Pedro Alves <palves@redhat.com>]

On 5/14/20 6:04 PM, Pedro Alves via Gdb-patches wrote:
> On 5/14/20 4:25 PM, Tom de Vries wrote:

> No objection to the splitting, but the issue above should really be
> fixed.  It's a systematic issue, and requiring a file split to address it
> seems like a heavy burden to me.  Maybe we should disable exec-file-mismatch
> detection from within gdb_target_cmd?  At least, until we have buildid detection.
> But that won't work on all hosts, either.  Like, this fixes align.exp for me:
> 
> diff --git c/gdb/testsuite/lib/gdbserver-support.exp w/gdb/testsuite/lib/gdbserver-support.exp
> index a2cc80f28d7..fc79f1d60c6 100644
> --- c/gdb/testsuite/lib/gdbserver-support.exp
> +++ w/gdb/testsuite/lib/gdbserver-support.exp
> @@ -52,7 +52,7 @@ proc gdb_target_cmd_ext { targetname serialport {additional_text ""} } {
>  
>      set serialport_re [string_to_regexp $serialport]
>      for {set i 1} {$i <= 3} {incr i} {
> -       send_gdb "target $targetname $serialport\n"
> +       send_gdb "with exec-file-mismatch off -- target $targetname $serialport\n"
>         gdb_expect 60 {
>             -re "A program is being debugged already.*ill it.*y or n. $" {
>                 send_gdb "y\n"
To sort of close the loop here -- I no longer think that this a good approach.

The exec-file-mismatch detection did actually find a bug in the testsuite:
 https://sourceware.org/pipermail/gdb-patches/2020-May/168680.html

I'm now aiming at fixing this in GDB, by making GDB validate the exec 
file by comparing build IDs instead of filenames:
  https://sourceware.org/pipermail/gdb-patches/2020-May/168718.html

Thanks,
Pedro Alves