------- Comment #19 from pcarlini at suse dot de 2006-10-31 17:01 ------- Created an attachment (id=12520) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12520&action=view) Draft patch using the _M_get_mutex idiom for _M_invalidate
This one, instead, using the _M_get_mutex idiom (similarly to pool_allocator, for example), should be able to deal with the _M_invalidate issue too. Agreed, in debug-mode performance isn't the main concern, still I'm a bit worried, because we have got one single mutex for the entire STL. But I have no idea how this problem can be solved at this stage, the ABI still frozen. Besides that general issue, the approach should work, maybe locking is needed also for some _Safe_sequence functions: in that case _Safe_iterator_base::_M_get_mutex would be still usable, being public, or, to be cleaner, an _M_get_mutex method could be added to _Safe_sequence_base too. In any case, I would rather wait for testcases before touching _Safe_sequence... Benjamin, what do you think? -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29496
