Re: gperf bug

[Morten Welinder]

   |>       register int key = hash (str, len);
   |>       if (key <= MAX_HASH_VALUE && key >= 0)

   |> Clearly, "key" should be unsigned and the wasteful test against 0 should
   |> be eliminated.

   A good compiler will do that anyway.

No way -- the implicit cast cannot be eliminated without an evaluation
of the range of hash, something that is way beyond what you should
expect from compilers.  If I had added the -G flag, it would not even
have been computable.

Having put in the right type for "key", I agree that the test is
likely to be eliminated -- with a warning from gcc, *sigh* -- by the
compiler.

Note, that the assignment may even overflow, in which case the currently
generated code will likely do the wrong thing.  (But it'll require a
large and strange set of keywords to actually make this happen.
Still, it's a bug that should be fixed.)

   |> (And perhaps gperf should use C++ comments in C++ mode.)

   What's wrong with using correct comment syntax?

It's not wrong, just out of style.

Morten