------- Additional Comments From pcarlini at suse dot de 2005-06-03 23:51 ------- Certainly, numbers from actual benchmarks would be useful. In order to make these comparisons easier, next weeks I will add to the v7-branch basic_string an alternate base-class implementation which avoids reference-counting (+ simple short-string optimization and needed changes to strings in exceptions classes). Will be useful also to start evaluating various options for forms of simulated move-semantics that should extend somewhat the range of application also to non-MT. All-in-all I would probably advice against the idea of opening a separate PR - the performance issue is *well* known, in the abstract.
-- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21334