This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: macrotab.c -Werror



Daniel Berlin <dberlin@dberlin.org> writes:
> It's an obvious false positive (!best will be true the first time through, 
> meaning the only time we check best_depth, it's already been set at 
> least once).
> 
> Here, you can't just initialize best_depth to 0, you have to initialize it 
> to either INT_MAX, or inclusion_depth (result).
> 
> Sucks.

You're going too fast.  Here's the whole loop, for the sake of
discussion:

  /* It's not us.  Try all our children, and return the lowest.  */
  {
    struct macro_source_file *child;
    struct macro_source_file *best = 0;
    int best_depth;

    for (child = source->includes; child; child = child->next_included)
      {
        struct macro_source_file *result
          = macro_lookup_inclusion (child, name);

        if (result)
          {
            int result_depth = inclusion_depth (result);

            if (! best || result_depth < best_depth)
              {
                best = result;
                best_depth = result_depth;
              }
          }
      }

The only reference to `best_depth''s value is in the right operand of
`||'.  That operand will never be evaluated unless `best' is non-zero.
But `best' is initially zero, and is only assigned along with
`best_depth'.  So `best_depth''s initial value is never used.  This
means:
- the original code is correct (although the compiler doesn't figure
  that out), and
- you can initialize it to anything you want, since its initial value
  is never used.

I don't actually know how many unnecessary initializations there are
in GDB to silence the compiler, but it's my impression that the
compiler's false positive rate for `var might be used uninitialized'
warnings is low enough that it's still a useful sanity check.  So I'm
happy to add a few unnecessary initializations.

What sucks (a bit) is that every one of those unnecessary
initializations does end up generating code --- if the compiler could
tell it was unnecessary, it wouldn't have printed the warning!


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]