RE: [PATCH 2/2 version 2] fix for gdb.reverse/finish-precsave.exp and gdb.reverse/finish-reverse.exp

[Carl Love <cel@us.ibm.com>]

Bruno, Tom:


On Thu, 2023-01-19 at 08:56 -0800, Carl Love wrote:
<snip>
> > 
> > If you have the test compiled, you can trigger the bug using
> > regular 
> > record, no need to use btrace specifically.
> 
> Thanks, I went back and tried the tailcall again.  I was able to get
> it
> to generate 2 fails on my laptop with just the X86 patch.  Not sure
> why
> I didn't see that before, maybe I had a typo??? 

I reviewed the full regression log on my X86 laptop and yup there are
the two errors for tailcall in the log.  I just missed them when I was
looking.  Argh.

>   I did note that git
> on my laptop seems to act a bit weird so maybe that was an issue??  I
> am having to double check the code changes after running git to make
> sure it applied the changes correctly.  Not sure why I didn't see the
> passes in the full regression suite either.  Anyway, I am looking to
> see if I can figure out what the regression issue is for this test.
> 
> The gdb.btrace/rn-dl-bind.exp test still runs fine with 9 passes.

I claim, the test case for gdb.btrace/tailcall.exp also needs fixing.

Without the patch, when you were in the main program at "answer += 1"
and did a reverse-next, gdb would go back thru the call to function bar
and thru the call to function foo stopping in main.  That doesn't seem
right to me as you have stepped back thru function foo and bar.  

With the patch the reverse-next only step back thru function bar and
stops in function foo at the call to bar, i.e. doesn't continue to step
back thru function bar.

The following is the tailcall source code with comments showing where
the reverse-step and reverse-next stop with and without the X86 patch
applied.  Hopefully the code below is a little easier to follow.
Initially, the reverse-step and reverse-next tests are executed from
the line "answer += 1;" in function main.


static __attribute__ ((noinline)) int
bar (void)
{
  return 42;
}                    <- reverse-step stops here (with no patch
                        and with X86 patch)

static __attribute__ ((noinline)) int
foo (void)
{
  return bar ();     <- reverse-next stops here (patch X86 applied)
}

int
main (void)
{
  int answer;

  answer = foo ();    <- reverse-next tops here (no patch applied, stepped
                         back thru both the bar and foo functions)
  answer += 1;        <- GDB is here, now issue the reverse-step
                         and reverse-next commands

  return answer;
}

As a result of the change in the reverse-next instruction, the expected
test results for the tailcall test need to be updated to reflect the
fact that gdb now stops in function foo not main.

--- a/gdb/testsuite/gdb.btrace/tailcall.exp
+++ b/gdb/testsuite/gdb.btrace/tailcall.exp
@@ -102,9 +102,9 @@ gdb_test "reverse-step" "\[^\r\n\]*main \\(\\) at tailcall.c
:37\r\n.*" \
     "reverse-step.2"
 gdb_test "next" "\[^\r\n\]*38.*" \
     "next.1"
-gdb_test "reverse-next" "\[^\r\n\]*main \\(\\) at tailcall.c:37\r\n.*" \
+gdb_test "reverse-next" "\[^\r\n\]*foo \\(\\) at tailcall.c:29\r\n.*" \
     "reverse-next.1"
-gdb_test "step" "\[^\r\n\]*foo \\(\\) at tailcall.c:29\r\n.*" \
+gdb_test "step" "\[^\r\n\]*bar \\(\\) at tailcall.c:24\r\n.*" \
     "step.1"
 gdb_test "finish" "\[^\r\n\]*main \\(\\) at tailcall.c:38\r\n.*" \
     "finish.2"

With this change, the tailcall.exp test now passes on my X86 laptop.
The PowerPC do not change since the test is not supported on PowerPC.

I will post an updated version of the X86 patch with the fixes to the
tailcall test.  It will be version 3.  There are no changes to the
PowerPC patch.

The gdb.btrace/rn-dl-bind.exp test passes with and without my patches. 
I still can not get this test to fail on my system.  

                    Carl