On Mon, May 11, 1998 at 06:43:20PM -0700, Martin Buchholz wrote:
My XEmacs packages are installed on a Linux machine.
Those packages are used from both Linux and from Solaris.
The Linux NFS server is notoriously slow.
AFAIK, this is fixed in the experimental kernel line. This is
irrelevant anyway :-)
Many of us on xemacs-beta are using XEmacs with all files on local
disk. In the real world, XEmacs is much more likely to be housed on
NFS-mounted filesystems on an overloaded network. We should work
under the assumption that the price of the current load-path layout is
TEN SECONDS in startup time. This is too expensive.
Three possibilities that I can see:
- cat the auto-autoload and the dumped-lisp files together, put
everything in one directory, reduce the load-path to two or three
directories. This breaks the packaging, unless you implement that
mainly with a forest of symbolic links. But afaik, symbolic links
aren't available in winblows.
- cache the directories content list in memory at startup, update it
automatically by checking for a directory date change when an
unknown file is asked for or an accessed cached file is absent. Then
fight with buggy NFS implementations
- same as 2, but cache on disk (aka, kpathsea). This also accelerates
startup.
My personal feeling is that anything different than the first solution
should wait for 21.1.
OG.