Re: [PATCH] gdb: use gdb_test_multiple in gdb_breakpoint

[Pedro Alves <pedro@palves.net>]

On 2023-01-10 3:50 p.m., Simon Marchi wrote:
> 
> 
> On 1/10/23 10:33, Pedro Alves wrote:
>> On 2023-01-05 4:28 p.m., Simon Marchi via Gdb-patches wrote:
>>
>>> ---
>>>  gdb/testsuite/lib/gdb.exp | 6 ++++++
>>>  1 file changed, 6 insertions(+)
>>>
>>> diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
>>> index e17eace4cb13..af538e5c8fbd 100644
>>> --- a/gdb/testsuite/lib/gdb.exp
>>> +++ b/gdb/testsuite/lib/gdb.exp
>>> @@ -657,6 +657,12 @@ proc gdb_breakpoint { linespec args } {
>>>                 send_gdb "$pending_response\n"
>>>                 exp_continue
>>>         }
>>> +       -re "$gdb_prompt $" {
>>> +           if { $print_fail } {
>>> +               fail $test_name
>>> +           }
>>> +           return 0
>>> +       }
>>>      }
>>
>> The other removed "-re" cases also considered $print_fail, so if their replacement
>> inside gdb_test_multiple is hit, they'll produce a FAIL.  Was that intended?
>> Should we instead add a "-nofail" option to gdb_test / gdb_test_multiple ?
> 
> Good point, this is a change in behavior.  Does this change cause you an
> unexpected FAIL in practice?  

No, I was just trying to catch up on patches a bit, and read your patch and noticed
that issue.  I was going to reply to the patch directly, but then saw the
follow up discussion and replied there (or rather, here).

> I was actually planning on removing that
> message / no-message option to gdb_breakpoint [1] to simplify it.  I
> don't see any use for the current behavior, I'd rather have it log a
> test result all the time.

I've always found the message/no-message API confusing.  It exists on
runto too, and maybe other procs.  

I think the intention of not issuing PASS by default, was that you can
use gdb_breakpoint to implement other procedures inside lib/gdb.exp.
If gdb_breakpoint starts issuing a PASS, then an implementation detail
of such procedures starts being visible, by ending up with two PASSes for
each call of the procedure that happens to use gdb_breakpoint, one for 
gdb_breakpoint, and one for the caller procedure proper.

I think it's the same reason many procedures in lib/gdb.exp were kept using
gdb_expect directly instead of being converted to gdb_test/gdb_test_multiple -- to
avoid the internal PASS/FAIL.

Pedro Alves

> I can kind of see when that would maybe be
> useful: if you wanted to set a breakpoint, and it could legitimately not
> work (you could do something with gdb_breakpoint's return value).  But
> you could also use the break command directly, like many places do
> already, so it's not really needed.  Anyway, all of this to say that I
> could fix what you pointed out by pruning / simplifying code rather than
> adding more.
> 
> Simon
> 
> [1] https://review.lttng.org/c/binutils-gdb/+/7158/6
>