eberard@grebyn.com (Ed Berard) (06/09/90)
The Ada Compiler Validation Capability (ACVC) contains a suite of test
programs used to determine the conformance of potential Ada compilers
with the syntax and semantics described in the Ada Language Reference
Manual (ALRM, ANSI/MIL STD 1815A). Within the test suite, there are
six major categories of test programs:
- Class A: These tests are supposed to compile and execute
with no problems.
- Class B: These tests are expected to fail compilation
- Class C: These tests are supposed to compile and execute
with no errors, but also do run-time checking
- Class D: These tests do capacity checking
- Class E: These tests check implementation-dependent features
and ambiguities
- Class L: These tests are supposed to fail at link time.
The general understanding is that ACVC tests are _not_ intended to be
used as benchmarks, nor are they intended to test the limits of an Ada
compiler. (There are a number of Ada Compiler Evaluation Capability
(ACEC) tests which claim to address these issues.)
Simply because a given Ada compiler successfully passes the latest
version of the ACVC, does not mean that the compiler contains no
errors. Nor does passing the ACVC mean _perfect_ compliance with the
syntax and semantics described in the ALRM. However, efforts are
constantly underway to improve the testing capabilities of the ACVC.
I am soliciting information from Ada compiler implementors and users
regarding that portion of the ACVC which tests generics -- a language
feature covered primarily in Chapter 12 of the ALRM. Specifically, I
am looking for ways to _improve_ the ACVC's coverage of generics. For
example, I would be interested in hearing about the following
situations:
- You are using a validated Ada compiler (validated under the
latest version of the ACVC) and you find that _valid_
(according to the ALRM) Ada source code does _not_ compile.
Equally important, the specific source code which does not
compile involves the syntax and semantics of generics.
- You are using a validated Ada compiler (validated under the
latest version of the ACVC) and you find that _invalid_
(according to the ALRM) Ada source code _does_ compile.
Equally important, the specific _invalid_ source code which
does compile involves the (potentially bogus) syntax and
semantics of generics.
- You are using a validated Ada compiler (validated under the
latest version of the ACVC) and you find that _valid_
(according to the ALRM) Ada source code compiles, but the
generated object code does _not_ behave correctly.
Specifically, the compiler generates code which does not
conform to the semantics described in the ALRM. Equally
important, the source code must specifically contain
generics and, to the best of your determination, the (bogus)
object code must be incorrectly implementing the semantics
of those generics.
- You are using a validated Ada compiler (validated under the
latest version of the ACVC) and you find that, although the
compiler compiles valid Ada source code, and the generated
object code conforms exactly to the semantics described in
the ALRM, there is some "unreasonable" constraint on the
behavior (or some other characteristic) of the generated
object code. Equally important, the source code must contain
generics, and, to the best of your determination, the
"unreasonable" constraint must be specifically tied to the
use of these generics.
- You have just noticed (or known about for some time) a
"gaping hole" in the way the ACVC tests generics. You may be
a (possibly frustrated) user who has been doing "write
arounds" to compensate for this problem. You may even be a
compiler vendor who has not yet had the time to bring this
to the attention of the Ada Validation Office (AVO).
For each report, I would find the following information most helpful:
1. A general description of the problem/deficiency
2. The specific Ada compiler(s) (vendor, version number, and
platform (hardware and software)) where you have found
the problem/deficiency.
3. The smallest (measured in lines of code) possible piece of
Ada source code which will reproduce the problem/deficiency.
4. Any additional information, comments, or suggestions you
may wish to make.
This information will be included in a report which will be placed in
the public domain. If you do not wish me to reveal some specific
information contained in your report, please indicate this.
This information will also be used to enhance the generics testing in
the next version of the ACVC, so please reply promptly. Thank you for
any time and consideration.
-- Ed Berard
Berard Software Engineering, Inc.
18620 Mateney Road
Germantown, Maryland 20874
Phone: (301) 353-9652
FAX: (301) 353-9272