This is the mail archive of the mailing list for the GCC 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: Gcc certification

I wrote:
> > The same is true of all other compilers that could conceivably be used
> > instead.  There are no bug-free compilers.

On Thu, Jan 20, 2005 at 06:01:47PM -0500, Robert Dewar wrote:
> Yes, absolutely, thanks Joe for clarifying this. Not only is the production
> of a certified compiler a daunting task, far beyond what we can do now, but
> we also lack clear formal language specifications to use as a starting point.
> Even certifying a compiler for a very small subset of a language like
> C++ or Ada is a huge task. There was a project at Boeing that claimed
> to be a proof of concept, but it was never certified. Furthermore it
> used interpretive tables for a lot of processing, claiming that the
> tables did not have to be certified since they were not code, but it
> is not clear that this would have been considered a valid approach.

You can't skip validating the tables.

The famous Pentium bug, which reportedly cost Intel 470 million dollars,
was due to an incorrectly generated table (the table was originally
correct, but the script to embed the table into a programmable logic array
had a bug).  Five positions in a lookup table used by the microcode for
the division instruction had the wrong value.

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