Sample clause3 tests

[Loren James Rittle <rittle at latour dot rsch dot comm dot mot dot com>]

Hi Brian,

As a part-time libstdc++-v3 maintainer, I read your recent article in
DDJ with interest (http://www.cs.clemson.edu/~malloy/projects/ddj.html).
You presented an interesting approach to test case generation/derivation.
It prompted me to install python and download your sample clause3 tests.

Here is some minor feedback.  By my reading of the ISO 14882 standard
section 3.7.3-2, only four functions are implicitly declared for use
by the new operator; the placement new forms are not in the presented
list.  (I only knew to look for this in the standard because I seemed
to remember a discussion was held on the gcc lists regarding this exact
point.)  Thus, g++ is correct to reject some of the cases you present
unless you explicitly include <new>.

Adjusting three of your 88 test cases for this issue gets gcc 3.1 down
to four clause3 test failures:

3.4.1-3
3.8-7 (still fails even after including <new>)
3.4.3-5
3.4.5-4

I do not yet know if those remaining failures are already covered by
our existing dejagnu test suite (but marked as failing).  With a quick
look, it appeared not.  (Note to CC'd: unless this falls under fair
use, I don't know if we can just copy them from the standard into our
gcc test suite.)

I should also mention that I saw only six clause3 failures with gcc
2.95.3 on i386-unknown-freebsd4.6 before I made any such adjustments
not the eight you reported with 2.95.2 on i386-*; which surely were
the same eight you reported with 3.0.4.  I am also fairly sure that we
didn't remove any bugs in this area on 2.95.X without also getting
them on 3.0.X.  Anyways, I only bring this to your attention since I
found it odd.

Thank you for making part of your test suite and results available.  I
await to learn how g++ 3.1 performs against your full test suite.

Regards,
Loren