Re: [PATCH v3 2/2] Fix an undefined behavior in record_line

[Andrew Burgess <andrew.burgess@embecosm.com>]

* Bernd Edlinger <bernd.edlinger@hotmail.de> [2020-03-27 04:50:29 +0100]:

> Additionally do not completely remove symbols
> at the same PC than the end marker, instead
> make them non-is-stmt breakpoints.
> 
> 2020-03-27  Bernd Edlinger  <bernd.edlinger@hotmail.de>
> 	* buildsym.c (record_line): Fix undefined behavior and preserve
> 	lines at eof.
> ---
>  gdb/buildsym.c | 34 ++++++++++++++++++----------------
>  1 file changed, 18 insertions(+), 16 deletions(-)
> 
> diff --git a/gdb/buildsym.c b/gdb/buildsym.c
> index 2d1e441..46c5bb1 100644
> --- a/gdb/buildsym.c
> +++ b/gdb/buildsym.c
> @@ -705,27 +705,29 @@ struct blockvector *
>  		      * sizeof (struct linetable_entry))));
>      }
>  
> -  /* Normally, we treat lines as unsorted.  But the end of sequence
> -     marker is special.  We sort line markers at the same PC by line
> -     number, so end of sequence markers (which have line == 0) appear
> -     first.  This is right if the marker ends the previous function,
> -     and there is no padding before the next function.  But it is
> -     wrong if the previous line was empty and we are now marking a
> -     switch to a different subfile.  We must leave the end of sequence
> -     marker at the end of this group of lines, not sort the empty line
> -     to after the marker.  The easiest way to accomplish this is to
> -     delete any empty lines from our table, if they are followed by
> -     end of sequence markers.  All we lose is the ability to set
> -     breakpoints at some lines which contain no instructions
> -     anyway.  */
> +  /* The end of sequence marker is special.  We need to reset the
> +     is_stmt flag on previous lines at the same PC, otherwise these
> +     lines may cause problems since they might be at the same address
> +     as the following function.  For instance suppose a function calls
> +     abort there is no reason to emit a ret after that point (no joke).
> +     So the label may be at the same address where the following
> +     function begins.  A similar problem appears if a label is at the
> +     same address where an inline function ends we cannot reliably tell
> +     if this is considered part of the inline function or the calling
> +     program or even the next inline function, so stack traces may
> +     give surprising results.  Expect gdb.cp/step-and-next-inline.exp
> +     to fail if these lines are not modified here.  */

Out of interest I tried reverting this patch and don't see any
failures in gdb.cp/step-and-next-inline.exp.  Could you expand on
which tests specifically you expect to see fail, and maybe which
version of GCC you're using?  I'm on 9.3.1.  It'll be Monday before I
can try my other machine which has a wider selection of compiler
versions.

I also don't understand what part of the previous behaviour was
undefined, could you help me to understand please.

Thanks,
Andrew



>    if (line == 0 && subfile->line_vector->nitems > 0)
>      {
> -      e = subfile->line_vector->item + subfile->line_vector->nitems - 1;
> -      while (subfile->line_vector->nitems > 0 && e->pc == pc)
> +      e = subfile->line_vector->item + subfile->line_vector->nitems;
> +      do
>  	{
>  	  e--;
> -	  subfile->line_vector->nitems--;
> +	  if (e->pc != pc || e->line == 0)
> +	    break;
> +	  e->is_stmt = 0;
>  	}
> +      while (e > subfile->line_vector->item);
>      }
>  
>    e = subfile->line_vector->item + subfile->line_vector->nitems++;
> -- 
> 1.9.1