RE: Our next GDB release (GDB version 6.8)

RE: Our next GDB release (GDB version 6.8)

[Pierre Muller]

  Joel,
  the main problem here is that
when you use the 'start' command
or set a breakpoint in 'main' function
and 'run', the behavior is not consistent
between ELF and other exe formats.
http://sourceware.org/ml/gdb-patches/2007-10/msg00361.html
  Currently GDB stops at the first line of real code
on ELF but on the call to __main
function.
  I added to the confusion because I tried to
argue that I was glad that it was that way,
and I proposed a testsuite solution to cope with that
problem that was rejected.
Daniel and Pedro convinced me that it would be much 
better to get the same behavior on all targets.

The problem is not restricted to i386 processor,
this is why Pedro proposed a patch with a new gdbarch
member. He only implemented the i386, but it
should be easy to extend this to other archs needed
this.

  I retested Pedro's patch tonight and 
it got
  174 more PASS
  175 less FAIL, 
no new FAIL appears
  1 more KFAIL (in fact a FAIL to KFAIL change).

  I really would like this one to go into
GDB 6.8

Pierre Muller
Pascal language maintainer

> -----Original Message-----
> From: Joel Brobecker [mailto:brobecker@adacore.com]
> Sent: Wednesday, January 30, 2008 7:04 PM
> To: Pierre Muller
> Cc: pedro_alves@portugalmail.pt
> Subject: Re: Our next GDB release (GDB version 6.8)
> 
> Pierre and Pedro,
> 
> >   I would really like to have the
> > call to __main fixed !
> 
> Would you mind reminding me what the actual problem is? I have a
> feeling
> at this point that there isn't a consensus on how to fix the problem
> yet
> and that could delay the release more than I'd like.  I would like to
> know more about the problem and its severity before we make a decision.
> 
> Thanks,
> --
> Joel

Re: Our next GDB release (GDB version 6.8)

[Joel Brobecker]

>   the main problem here is that when you use the 'start' command or
>   set a breakpoint in 'main' function and 'run', the behavior is not
>   consistent between ELF and other exe formats.

Got it: http://www.sourceware.org/ml/gdb/2006-11/msg00140.html.
Thanks.

> I really would like this one to go into GDB 6.8

In my opinion, this is not a critical issue. IIUC, this is not a regression,
and the only issue is that, for the "main", the breakpoint will be placed
at a location such that the debugger will stop at the open curly brace.
So the user will need an extra "next".

I did notice that the fix gets rid of a lot of FAILs though.

If it was just me, I would categorize this as non release-critical.
But the rest of the group might disagree, so I'll defer to them.

-- 
Joel

Re: Our next GDB release (GDB version 6.8)

[Pedro Alves]

Joel Brobecker wrote:
> In my opinion, this is not a critical issue. IIUC, this is not a regression,
> and the only issue is that, for the "main", the breakpoint will be placed
> at a location such that the debugger will stop at the open curly brace.
> So the user will need an extra "next".
> 

That and the fact that global c++ ctors will only run after "next".
That's what's most different, and surprising.

> I did notice that the fix gets rid of a lot of FAILs though.
> 

This happens, because a lot of tests fail because runto_main or
similar fails.

> If it was just me, I would categorize this as non release-critical.
> But the rest of the group might disagree, so I'll defer to them.
> 

I agree.  I'd like this to go in, of course, but there's no rush
from me.

-- 
Pedro Alves