Wildcard-specs

[ERCF08 Bob Gray]

In article <4162 at brl-smoke.ARPA> lacasse at RAND-UNIX.arpa writes:
>
>If you have all day, you can use:
>    find . -exec grep string {} \;

You would be faster with
	grep string `find . -print`
provided that there weren't too many filenames. There is a
limit built into your system on the number of parameters you
can pass in an exec().

It has always seemed odd to me that a system with a
hierarchical file system didn't have some pattern which
recursively matched to sub directories and the files in
them. It would be very nice to do something like

	vi **.mk

to edit all the makefiles in a particular set of directories.

In most cases the pattern matching of
e.g.
	ls -l **
has the same meaning as
	ls -l `find . -print`

The problems begin when you attempt to make the "**"
wildcard as general as the other metacharacters. The pattern
"**x**"  means "find any file or directory which has the
letter x in its name. match to either the filename or
to ALL the files in the sub-tree if it is a directory.
i.e.
	**xyz**
should match to
	abcxyz.x
	abc/abcxyz.y
	abc/def/xxyzz/a.c
	abc/def/xxyzz/b.c
	abc/def/xxyzz/d.c
	abc/def/xxyzz/e.c

I started to implement "**" pattern matching in the Bourbe
shell a few years ago. Getting it to match with example 1.
was fairly trivial. The more general "**xyz**" case caused
an explosion in the number of filename matches before
sorting and elimination of duplicates. It was very easy to
run out of space on the machine I was working on in those days.

If I dig the code out of the archives, what sould happen in
a command like
	cp **.c x

where x is a directory. Should the shell quietly make new
directories or should it issue seperate cp commands for each
target directory, then let cp complain if they don't exist.
	Bob Gray.
	ERCC.