[gcov] Avoid race conditions in parallel make

[Jan Hubicka]

> Jan Hubicka wrote:
> 
> >! 	      /* If everything done correctly, the summaries should be
> >! 	         computed equal for each module.  */
> >! 	      if (last_prg.runs
> >! #ifdef TARGET_HAS_F_SETLKW
> >! 		  && last_prg.runs == local_prg.runs
> >! #endif
> >! 		  && memcmp (&last_prg, &local_prg, sizeof (last_prg)))
> >  		{
> >! #ifdef TARGET_HAS_F_SETLKW
> >! 		  fprintf (stderr, "profiling:%s:Invocation mismatch\n\
> >! Probably some files were removed\n",
> >  			   ptr->filename);
> >+ #else
> >+ 		  fprintf (stderr, "profiling:%s:Invocation mismatch\n\
> >+ Probably some files were removed or parallel race happent because 
> >libgcc\n\
> >+ is compiled without file locking support.\n",
> >+ 			   ptr->filename);
> >+ #endif
> >  		  local_prg.runs = 0;
> >  		}
> >  	      else
> >! 		memcpy (&last_prg, &local_prg, sizeof (last_prg));
> I don't think this is right.
> 
> Firstly, that extra check for last_prg.runs == local_prg.runs is 
> superfluous,
> the memcmp will pick it up anyway. Secondly, this will detect a difference
It is not - it is in reverse.
In the case someone else is updating the profiles in parallels, the
number of runs and summaries won't match, but still I must not abort.
> when one invocation of a program overtakes another, during the update
> of the gcov files, even though that kind of situation should be dealt with.
Yes, that is why I put the runs check there.

Honza
> 
> nathan
> -- 
> Nathan Sidwell    ::   http://www.codesourcery.com   ::     CodeSourcery LLC
>          The voices in my head said this was stupid too
> nathan@codesourcery.com : http://www.cs.bris.ac.uk/~nathan/ : nathan@acm.org
>