Brad Balfour said:

  > If you, the maker of the compiler, want to ensure it's
  > correctness, then all obtain a copy of the ACVC, delete those test
  > that don't make sense for your (subset) compiler, and run the test
  > suite. You can even provide these results to your customer upon
  > request.

  > As long as we are not talking about the US DoD -- and we are not
  > -- then validation is not a requirement at all. Let's not let this
  > impede our progress into new commercial domains.

   Why not obtain the ACVC, run all the tests on your cross-compiler,
and ignore any that can not be linked due to capacity reasons, or
which won't execute due to the lack of key hardware features.

   Then apply for validation, you should get it.

   AI-325 has migrated into chapter 1, but it has always been part of
validation policy.  If the hardware don't support it, then don't test
for it.

   We had a huge hoorah around here once about the floating point
tests and 80286 machines with no '287 chip and a contract requirement
for a validated Ada compiler.  (On those particular machines, there
was no way to easily install a software trap.)  We finally got
everyone involved convinced that the "right" answer was that any PC
Ada compiler qualified, and that people who wanted to use floating
point needed the optional version with the 287 chip.


                                        Robert I. Eachus

with Standard_Disclaimer;
use  Standard_Disclaimer;
function Message (Text: in Clever_Ideas) return Better_Ideas is...