Doug Gregor <dgregor@cs.indiana.edu> writes:
On Jul 13, 2004, at 5:49 AM, David Abrahams wrote:
Synge Todo <wistaria@comp-phys.org> writes: It appears that the old input iterator requirements require that *r++ returns value_type instead of a proxy. However, *r++ must be a proxy if the iterator is writable. Since there's no way to detect iterator writability reliably, I'm going to have to make *r++ return value_type whenever the specified iterator category is strictly std::input_iterator_tag.
I'll let everyone know when it's checked in
FYI, the operations_test.cpp test of the filesystem lib fails unless the return type of *r++ is value_type (where r is a directory_iterator). The problematic line is, of course:
BOOST_TEST( (*dir_itr++).leaf() == "d2" );
directory_iterator uses single_pass_traversal_tag, not std::input_iterator_tag. Granted, there's a big comment in that same block of code (near line 100 of boost/filesystem/operations.hpp) that indicates how unsure the author was about proxies :)
Yeah, I decided that we'd only make a writable proxy for *r++ if the facade's CategoryOrTraversal parameter is convertible to std::output_iterator_tag. It's the best thing I can think of, since I expect that building output iterators with iterator_facade is rare. Tests seem to be passing now; I'll check things in shortly. -- Dave Abrahams Boost Consulting http://www.boost-consulting.com