On Thu, 02 Sep 2004 10:43:36 +0100, Daniel James <daniel@calamity.org.uk> wrote:
I've been trying out your new version, and I found that it was fairly easy to run out of indicies when using it repeatedy. Looking at the code, I think it should be possible to only generate one value per encoding. Basically my idea is that instead of using:
friend sizer<VOLATILE> encode_value(encode_counter<Index>*);
You could use:
template<unsigned EncodingId, unsigned Index = 0> struct value_pair { typedef value_pair<EncodingId, Index+1> next; };
friend sizer<VOLATILE> encode_value(value_pair<EncodingId, Index>*);
The existing compile time variable mechanism would just be used to generate the increasing EncodingId's - one per typedef. I included the next member so that the value_pair could be passed by type, replacing the existing Index parameter. Then whenever you currently use Index+1, you can use 'typename ValuePair::next'.
I guess this is a little obvious, so sorry if you're already doing something like it. If it doesn't make sense I can put together an implementation.
Daniel
In the current version of typeof that you have, I made a very quick and dirty workaround for a MSVC ETI problem. The problem arose from typeof being used inside a template class, and caused the starting index values to be garbled. As a quick workaround, every time BOOST_TYPEOF is used, the index is incremented by 100, meaning you can only use BOOST_TYPEOF 10 times (with the current BOOST_MAX_TYPEOF_COUNTER) When I analyzed the problem later, I discovered that all that is needed is to set the starting index to 100, and the problem goes away. I fixed the problem and resubmitted it to boost/files. In this implementation, the index is greedy, meaning that a type with depth of 2 will only occupy 2 indices. If you still encounter problems, try to increase BOOST_MAX_TYPEOF_COUNTER I also made some additional fixes to make typeof compile in release (VC 6.5) (Requires Multithreaded DLL compile condition in order to not ICE...) Good luck. Thank you for the feedback.
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost