bvs@light.uucp (Bakul Shah) (05/03/90)
The other day as i was applying a zillion patches to the perl3
tree, which went pretty smoothly, by the way, it struck me that a
patchfile is functionally sooo... close to a sharfile! Patching
with patch is not only entertaining, it is also faster than
grinding through unshar, sh, cat/sed, wc and a tmpfile or two.
Patch runs on more machines than shell and everyone can get patch
sources. It is also all you need for upgrades and patch is safer
than using sh (and can be made safer -- yet another option to
touch files only in the current tree).
What we need is a version of diff such that
diff -c /dev/null source-tree
will produce an initial `patchfile' which when fed to patch will
reproduce the source-tree. Better still would be
mkpatch [-f old-tree] [-s patchfile-size] new-tree
which creates a initial patchfile or one by diff'ing new-tree
against old-tree, creates a patchlevel.h if one does not exist,
and may be, handles file permissions, safely transports
unprintable chars, (though these are is perhaps best handled in a
makefile) etc. Patch creates directories as needed but it may
have to be extended to handle file permissions and unprintable
chars.
With a few more changes, such as interface to sccs/rcs or a
separate checksum file (ala zoo) to be checked by patch, we can
have a pretty nifty software source distribution system! Now
*that* would be worth standardizing.
Comments? Any takers?
-- Bakul Shah
bvs@BitBlocks.COM
..!{ames,att,decwrl,pyramid,sun,uunet}!amdcad!light!bvs