Re: Ada Core Technologies and Ada95 Standards

[eachus@spectre.mitre.org (Robert I. Eachus)]

    (I don't know if this discussion is being followed by many
non-participants, but it certainly is on topic.)

    I think Ken is missing the point Robert is making--not that ACVC
type testing of Ada compilers is in need of improvement, but that any
more emphasis on testing _instead_of_other_means_of_improving_quality_
would result in decreased compiler quality.

    I would be surprised if ACT performs less than a MIPS-century of
ACVC and regression testing each year.  Also, from experience, I
suspect that tests that execise the code generator (i.e. other than
ACVC B or L tests) are run much more often that those tests which are
expected to produce error messages.  (Even if you only changed the
front end, you want to test against most code generators, but running
the B-tests once is probably sufficient.)

    I doubt that the F-22 program tests at this level.  Speaking from
experience, the testing on large programs like this are in the
MIPS-month per year range if we are lucky.  That is still one HUGE
lump of testing.  However, the F-22 program probably spends much more
time on formal proofs of correctness--not for the whole system, but
for particular algorithms, like the control laws.  In the compiler
business, a lot of this work is done by tools instead.  I don't know
how much of this ACT does, but I know that, for example, DDC-I uses
VDM tools to formally prove that components meet requirements.  This
approach is very viable on compilers because the effort put into
making assertions about, say parse trees, is used in testing all code
that modifies those trees.  (And in any good Ada compiler, that
includes a lot, usually including but not limited to, parser, semantic
analysis, high and low level optimizers, and hardware specific
modules.) 


--

					Robert I. Eachus

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