------- Comment #2 from pcarlini at suse dot de 2007-06-11 15:25 ------- Hi Benjamin. If you ask me, this issue is INVALID: it is *not* supported building the library normally and then passing -D_GLIBCXX_FULLY_DYNAMIC_STRING on the command line, and never was! Simply because basic_string is exported, used in exported bits of locale, you know what I mean... As usual in such cases (e.g., changing locale model), the user has to configure with --enable-fully-dynamic-string and rebuild. In fact, note that the macro in the patch has the conditional wrapped in #ifndef _GLIBCXX_FULLY_DYNAMIC_STRING, code which at some point was supposed to make a difference only when a *normal* string was involved (in fact it doesn't, and that's why I removed it, the check is redundant): the fact that adding back that code avoids the segmentation fault is only because, by chance, this particular combination of exported normal string + inline fully dynamic string works apparently better, but nobody knows what would happen in other much more complex situations... That said, if you really feel strongly about adding back that conditional, it's up to you, but I would suggest at least adding a clear comment, summarizing the above.
-- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32284
