tyler@gen1.UUCP (Tyler IVANCO) (11/18/86)
I've recently been partially successful in porting gnu-emacs 17.64 to a Nat Semi ICM 3216 system under BSD 4.2. However, I've encountered a few problems (more like nuisances). The first involves the gnu-emacs notion of an address. It seems that the gnu routines are referenced through a 24 bit address. The address is combined with other information bits so that the total union is a 32 bit object. This works fine for those whose compilers generate routine calls via virtual addresses (e.g. jsr <address>) but does not work at all with systems that call via a module table pointers (i.e. cxp <module table entry>). The module table (something whose time has come), requires a full 32 bits. Thus the gnu notion of a 32 bit integer in union with a 24 bit address/other information is inconsistant with the processor/compiler. Fortunately a jsr directive was available in the compiler. Unfortunately, the UNIX signal system required module table entry values rather than virtual address values. Thus a kludge was required to coerce the virtual address to module entries for signals. Has anyone been able to lengthen the gnu-emacs addressing union so that the use of the module entry is possible? The second question involves the virtual size of emacs. After running for a short while, I find that the virtual space has grown in excess of 3-4 Mbytes. Using gosling emacs, I have yet to use more that 800K of virtual space. Have I screwed something up in my hacking or is this characteristic of gnu-emacs? My final question involves the shell manager. I find that my commands are always re-echoed unlike gosling emacs. Again I suspect one of my hacks. Comments? Tyler Ivanco BITNET: FS300022@YUSOL ...utzoo!yetti!gen1!tyler ...seismo!mnetor!/