Re: auto_ptr / iostream bugs

[Benjamin Kosnik <bkoz at redhat dot com>]

> In any case, for auto_ptr, I did not really think about adding the
> 'negative' test to the test suite, but I should have. Auto_ptr was
> designed to NOT work under certain circumstances and it would be a
> good idea to test them. For the other case, however it appeared that
> there was already a test that should have caught the typo so I do not
> know why it didn't. I will investigate it further.

..known bug, more likely. There is now an auto_ptr_neg.cc file for
auto_ptr negative tests.

> For the iostreams stuff, I will go back and redo the patch into the
> standard form. It has been suggested that it was too large. I know it
> involved several files, but it was basically triggered all by one
> thing. Which would be the preference, individual patches - one per
> file - or one email with a bunch of attachments?

I've integrated the more obvious fixes into gcc-3.2. The individual
patches should not require an assignment, per se, although doing one is
a good idea. (And will, as Phil said, be necessary for the testsuite
work.)

http://gcc.gnu.org/ml/gcc-patches/2002-07/msg00225.html

I didn't get a chance to do some of the io types, as noted. I'd
appreciate another patch for those, in much the same style as what was
posted, if you are feeling generous with your time.

In additon, to add testsuite entries for this, you just need to add the
custom char_traits class, and test only for compilation, not linkage.
(Several of the 22_locale tests do this, for the same reason.)

Therefore, to do the testsuites, you'd need to add explicit
instantiations exercising the custom traits_type to 

27_io/istream.cc
27_io/ostream.cc
27_io/streambuf.cc
27_io/filebuf.cc
27_io/stringbuf.cc

etc. I know istream/ostream are ready for this, but perhaps some of
these files need to be split into streambuf_members.cc and streambuf.cc,
etc.

Perhaps this is too confusing. If so, I can work on this when I get
back.

Thanks, this is very useful.

-benjamin