Aidan Kehoe writes: > In local testing, I can reproduce your bug; and it seems that Stephen’s > patch of http://mid.gmane.org/87hcixwkh4.fsf＠uwakimon.sk.tsukuba.ac.jp , > which he has yet to commit, eliminates it. We’d love confirmation of this > from you; despite Stephen’s ‘do not apply this patch’ in his message, the > change is easy to apply. Now that I've got a test that reproduces, I'll apply it tomorrow. Note that the reason for third parties not to apply is that I intend to make a whitespace change, and their CVS checkouts will have conflicts due to that since the posted patch ignores whitespace. However, there may very well be another bug lurking, as the backtrace I was looking at was in a function that a pure ASCII search pattern should never reach.

(If somebody cares to look, the function is search.c, simple_search(); if only ASCII is in use, boyer_moore() should always be used instead.) > > Ar an t-ochtú lá de mí na Nollaig, scríobh Klaus Reim: > > > Hi! > > > > I encountered a fatal error during an isearch-forward > > within a several 10000 lines buffer. > > Eventually it turned out that "size does not matter": > > I was able to boil the problem down to a mere 3-bytes buffer > > containing the letters "IAI". > > If you place point just before the 'A', and then > > do a "C-s C-w" sequence, xemacs dies from: Hm, interesting. > > > > Fatal error: assertion failed, file search.c, line 1487, > > (this_pos) > ((Bytebpos) 1) && this_pos <= ((buf)->text->z + 0) > > > > instead of marking the "AI". > > This occurs for example in today's cvs-head. > > To be more precise, the assertion triggers for > > CVS date-tag "-D 2007-08-27" and later, while it does not for "-D 2007-08-26" > > Among the few files changed in that interval, it turned out > > that revision 1.9 of "lisp/mule/latin.el" causes the problem > > (or maybe makes an older problem evident). > > I built my version of xemacs --with-mule. This absolutely cannot happen without Mule, since that assert is only enabled with Mule. > > > > Find below my system info (saw the same effect on > > a redhat system as well, with the same build configuration). > > > > regards, > > Klaus