From: Bruno H. <br...@cl...> - 2005-02-24 17:59:12
|
Sam wrote: > I even thought that this might make the clisp driver unnecessary: I would be satisfied if there was no need to hardwire a pathname in the clisp driver. > make the clisp driver unnecessary ... > -B/-M defaults -- can be extracted from exec filename (use -M "" for > bootstrap) What would be the purpose of this? We have a clear concept of what is a linking kit: an executable plus a memory image alongside (plus some other files). The documented way to run these is through "clisp-link run" or the driver. I see no purpose of allowing to run the lisp.run directly, since it *will* mess up with symlinks and hard links. (CMUCL and SBCL have mechanisms similar to what you propose, and their automatic determination of the memory image doesn't allow "sbcl" to be in $prefix/bin. To work around this, I have to add a tiny driver script in $prefix/bin. And so we're back to the original problem.) > -K - obsoleted by dynamic modules So far, dynamic modules don't obsolete anything because they are not portable. > /usr/bin/clisp is just a symlink to /usr/lib/clisp/base/lisp.run > (and find_executable() actually calls realpath() so this is not a problem) I don't see how this could work on Windows on VFAT filesystems (which have no concept of symlinks or hardlinks) or on BeOS filesystems (which have no concept of symlinks). Bruno |