Eric Niebler wrote:
Inline ...
Martin Wille wrote:
I now get these errors for gcc 2.95.3:
regress.cpp:88: invalid operands `const boost::foreach_detail_::auto_any_base &' and `int' to binary `operator !='
As I mentioned, gcc 2.95 accepts BOOST_FOREACH in some onstructs but not others. In particular, the compiler chokes when BOOST_FOREACH is in a function template, like above. Simpler uses are fine. I'd gladly accept a patch, but I'm not inclined to waste much more of my time on this ancient compiler.
I wouldn't have supported it in the first place :)
gcc-4 without #defines:
/boost/head-regression/boost/boost/iterator/counting_iterator.hpp:32: error: '(bool)(boost::is_integral<char>::value)' is not a valid template argument for type 'bool' because it is a non-constant expression ../../../boost/foreach.hpp:281: error: '(size_t)((-1))' is not a valid template argument for type 'unsigned int' because it is a non-constant expression
Whoa. This is some pretty basic, legal C++ that gcc-4 is choking on. Has gcc-4 been released yet? If not, I'm going to wait for it to stabilize a bit.
No it hasn't been released, yet.
gcc 3.2.3 without defines:
regress.cpp: In function `std::vector<int, std::allocator<int> > to_vector_foreach_byval(Range&) [with Range = int[5]]': regress.cpp:132: instantiated from here regress.cpp:88: no match for `bool ? boost::foreach_detail_::rvalue_probe : int[5]' operator
I'm guessing that defining BOOST_FOREACH_NO_CONST_RVALUE_DETECTION makes this go away? That's probably the right fix. It looks like gcc-3.2 is choking on the fancy rvalue detection stuff. I'll fix this.
Well, I'd consider gcc 3.2 not worth supporting. However, since you asked for gcc 2.95 in your first message, I thought you might be interested in the results for gcc 3.2, too. Yes, the error goes away with BOOST_FOREACH_NO_CONST_RVALUE_DETECTION. (The very same applies to gcc 3.1 and gcc 3.0.4) Regards, m