JV> [discussion moved to beta] JV> Martin Buchholz <martin(a)xemacs.org&gt; writes: JV> I am wondering whether this might have been intentional originally so JV> you don't forget to GCPRO when passing it to functions. I'm pretty sure it's because it can't be maximally efficiently implemented unless the compiler supports inline functions. GCPROing was annoying before, and will be annoying after, until I implement conservative gc for the stack, or someone beats me to it. BEFORE: Lisp_Object foo; Lisp_FOO *c_foo = alloc_lrecord_type (Lisp_FOO, &lrecord_FOO); XSETFOO (foo, c_foo); GCPRO1 (foo); bar (foo); UNGCPRO (foo); AFTER: Lisp_FOO *c_foo = alloc_lrecord_type (Lisp_FOO, &lrecord_FOO); Lisp_Object foo = Lisp_FOO_to_Object (c_foo); GCPRO1 (foo); bar (foo); UNGCPRO (foo); It's true that there will be the urge to just do bar (Lisp_FOO_to_Object (alloc_lcrecord_type (Lisp_FOO, &lrecord_FOO))); which will lose horribly. But someone will submit a patch out of the blue to implement conservative gc for the stack, right? It happened for mule support on Windows (thanks Tomonori!). and then our C code will be pure beauty. Martin