Re: 21.2-b19 crash on Solaris 2.7

[ Martin, could you have a look at this? I think this might be related to your changes to mapcar1 in December'98, but I'm not smart enough to figure it out... ] Georg Nikodym <georgn(a)canada.sun.com&gt; writes: We have seen this a couple of times before with different compilers on different platforms: Simon: http://www.xemacs.org/list-archives/xemacs-beta/9901/msg00718.html Colin: http://www.xemacs.org/list-archives/xemacs-beta/9906/msg00076.html Raymond: http://www.xemacs.org/list-archives/xemacs-beta/9906/msg00123.html http://www.xemacs.org/list-archives/xemacs-beta/9910/msg00046.html Cameron: http://www.xemacs.org/list-archives/xemacs-beta/9908/msg00018.html Georg: http://www.xemacs.org/list-archives/xemacs-beta/9909/msg00235.html The pattern I see is something like: - execute_optimized_program calls Fmapcar or Fmapconcat via Ffnucall - the function passed to Fmap[concat|car] either is or at least involves char-to-string - the SEQ is a string. I looked at this some time ago and made some wild guesses related to GC and Martin's 'Marzilla' changes: http://www.xemacs.org/list-archives/xemacs-beta/9906/msg00191.html looking at this again I'm not so sure this is really a GC problem as execute_optimized_program seems to do all the necessary GCPROing (as Hrvoje pointed out). Maybe there is some subtle bug in the interaction of Fchar_to_string and the (new) string code in mapcar1. I'll look at this again. If anyone sees this again, please send a LISP backtrace in addition to the C backtrace and if possible save the core for further examination. Naturally I prefer backtraces from debug builds... Gunnar