On Mon, Sep 11, 2006 at 06:43:19PM +0200, Matthias Hopf wrote: > Well, not exactly trial-and-error. > They probably have design specs, but what is actually delivered in > silicon is typically quite a bit off. Some things turn out not to be > implementable, some trigger a slow path, some things are buggy.
Ah, now that again sounds a bit more realistic. > You have to work around in the driver, and I assume that approx. 30-50% > of the code is about workarounds. This is only an educated guess, so > don't take my words for granted. > > Of course some errors are only found by trial-and-error, but that is the > case in the whole software industry. Even if you use formal methods, in > that case the driver might do exactly what you specified, but what you > specified is not necessarily what you actually wanted... Yes, this is true for (almost) the whole software industry. But you can partition the whole software industrie into two groups: The one that has so much clue that they update their specs or at least document the problems to prevent walking into their own trap again and the one that has not. --- From implementation reviews I must admit that the second group might be significantly larger... Robert -- Robert Schiele Dipl.-Wirtsch.informatiker mailto:[EMAIL PROTECTED] "Quidquid latine dictum sit, altum sonatur."
pgpvZZZjo7RXo.pgp
Description: PGP signature
