Robert Ramey wrote:
Bill Lear wrote:
I'm not sure I need to use a cache of strings when saving, in fact, I think not, but I went ahead and more-or-less followed your instructions.
I think I was thinking along the lines of the following template<Archive> void save(Archive &ar, ll_string & l, const unsigned int version) const { std::setstd::string::iterator it = m_cache.find(l); if(m_cache.end() == it) m_cache.insert(l); // serialize a pointer to the cached value ar << & (* it); // or ar << it - serializing an iterator!!! that's a first } void load(Archive &ar, ll_string & l, const unsigned int version){ std::string * t; ar >> t; // warning or emitted by serialization library - try & or cast. // copy de-serialized string to "real" destination l = *t; } Thus, all the strings what are equal to each other would have all have the same pointer. Serialization tracking ensures that only one copy of the same object is in the archive. So when you load, all the instances have thier contents shared. This could make the file much smaller and faster to load. For example suppose were going to serialize the bible. (King James version). and we have it stored as a long list of words. In this system all words would be stored only once. + one small integer each time the word is stored again. Given a guess of 10,000 different words used in this 250,000 ? word work - the archive would be a lot smaller. Also it might be considerably faster to load. I'm not sure if this would really work - its just an idea you might want to take this idea with a grain of salt. Robert Ramey