Didier Verna <verna(a)inf.enst.fr&gt; writes: More importantly they make it close impossible to use the debugger. (Here debugger is defined as me+gdb). I wish the same sometimes too. Some parts of XEmacs are a real good argument for C++ (at least in theory, and some compilers are getting pretty close to that now): Modern compilers (gcc 2.95 with --new-abi) have all these - One word overhead on RTTI[1] - Templates to get rid of those huge macros in lrecord.h etc - function pointer based OO replaced by language syntax support OO. - Table based exception handling (basically the same as our specl* stuff, I think). Given that you can do a simple lisp in #!/bin/sh it is no surprise that you can do it in C++, but I think that a bit of language support couldn't hurt the readability of the code. Dumping a C++ executable is going to be a pain though. I also am not sure whether allows smarter elisp implementation than we have now (i.e. one with a copying garbage collector and real closures). Jan Footnotes: [1] If we are going to release 21.2 with LRECORD_CONS then we really should consider reducing this overhead, 50% is simply too much.