Is there a way to remove files that don't match a given string?  I
don't think rm has such a flag.

crude, but effective:
   rm `ls ./* |grep -v .tmp`

There's also 'rm -i *', and answer 'n' to any '*.tmp' files.  <grin>

If you're using zsh, I would recommand that you use it's
extended globbing:

setopt extendedglob # to put in your ~/.zshrc

Then, simply do:

rm ./^*.tmp

If you're using ksh, you can do

rm ./!(*.tmp)

If you're using bash, you could do the same after a shopt -s
extglob, and with zsh after setopt kshglob

If not, you can do:

find . \! -name '.' -prune \! -name '.*' \! -name '*.tmp' \
  -exec rm {} \;

--
Stephane

You need the -d option of ls. It will remove files that don't
have "tmp" in their name (the "." matches any character, so is
useless here). There will also be problems with filenames with
SPC, TAB, NL, *, ?, [, \ characters.

--
Steph

for i in ./*; do
  case $i in
    *.tmp) :;;
    *) rm "$i";;
  esac
done

Or, to build a list of files and remove them all at once:
shift $#
for i in ./*; do
  case $i in
    *.tmp) :;;
    *) set "$i" ${1+"$@"};;
  esac
done
if [ "$#" != 0 ]; then rm "$@"; fi

These should work in any Bourne-compatible shell.

paul

Not in zsh in sh or ksh compatibility mode (in sh compatibility
mode word splitting and filename generation is forced in
${1+"$@"} case, that's actually not really a bug, but just how
the nesting of parameter expansion operators works in zsh).

The common work around for this is to use:

case ${ZSH_VERSION+z} in z) alias -g '${1+"$@"}' '"$@"';; esac

Or you can stick to POSIX conformance and use "$@".

Note that if there's no file in the current directory, you'll
get an error from "rm" saying that it can't remove a file named
"*".

--
Stephane

If "compatibility mode" isn't actually compatible, I don't see how
that can be anything but a bug.  The mode exists for no other reason
than to produce the same behavior as sh, and in this case, it fails to
do that.

Thanks, I forgot about those bits.

paul

You could also use the "find" command like this:
find . \( -name '*' -a ! -name '*.tmp' \) -print
find . ! -name '*.tmp' -print

If the command is ok then just add:  -exec rm {} \;
at the end of the command
find . ! -name '*.tmp' -print -exec rm {} \;

Et voila the xxx.tmp files will be gone...

I hope that will help you...

Well, let's say it's an emulation mode. zsh has a csh, ksh and
sh emulation mode. They wont exactly mimic csh, ksh or sh
because zsh syntax is very different in some ways. For instance
csh mode is more a csh look-a-bit-like mode as zsh is definitely
a Bourne-type shell.

Note that in other shells, ${1+"$@"} is not always consistent
from one shell to another:

$ sh -c 'printf "<%s>\n" "${1+"$@"}"' - 'a b'
<a>
<b>
$ ksh -c 'printf "<%s>\n" "${1+"$@"}"' - 'a b'
<a>
<b>
$ bash -c 'printf "<%s>\n" "${1+"$@"}"' - 'a b'
<a b>

--
Stephane

    It make no sense to quote ${1+"$@"}.

    I tried this script (in a file and called in the normal manner)
    with each shell (e.g., pdksh ./xx.sh):

set -- "a x" b c
for p in  ${1+"$@"}
do
   echo "<$p>"
done

    It demonstrated the expected and desired behaviour. The output for
    bash, ash, zsh, pdksh, ksh93, ksh98, FreeBSD's sh and a Bourne
    shell (SunOS 4.1) was:

<a x>
<b>
<c>

    However, when I used "${1+"$@"}" (quoted), ksh (88 and 93) and sh
    on FreeBSD, produced:

<a>
<x>
<b>
<c>

--
    Chris F.A. Johnson                  http://cfaj.freeshell.org/shell
    ===================================================================
    My code (if any) in this post is copyright 2004, Chris F.A. Johnson
    and may be copied under the terms of the GNU General Public License

That's because of your quoting "${1+"$@"}" :

sh -c 'printf "<%s>\n" ${1+"$@"}' - 'a b'

bash -c 'printf "<%s>\n" ${1+"$@"}' - 'a b'

both give consistent results:
<a b>

.
.... [snipped]
.

this would also try to remove subdirectories in the current directory
as you did not qualify '-type f' following '-prune' and also
any hidden files ending in .tmp would be removed too.

/bin/find . \! -name . -prune -type f \
            \! -name '*.tmp' \! -name '.*.tmp' \
-exec rm {} \;

for file in ./* ./.*;do
 case $file in
  *.tmp) :;;
  *) [ -f "$file" ] && [ ! -L "$file" ] && rm "$file";;
 esac
done

Bzzzzt!  Thank you for playing.
'.' is *NOT* a wild-card character in file-name globbing.  

But it is special to grep.  You probably meant: grep -v '\.tmp'
But that still leaves the other problems Stephane mentioned.

paul

Agreed for the directories. Then, you can add either "-f" (to
ignore the errors) or -rf to actually remove the dirs to rm.

But  .foo.tmp wouldn't be removed because of the \! -name '.*'
(added to avoid removing dotfiles, as in the other shell
globbing solutions I provided in the part you snipped).

With that one and with a POSIX conformant find, dot files not
ending in .tmp would be removed.

And with a non-POSIX conformant find such as GNU find, a file
named ".tmp" would be removed.

Also, it doesn't remove non-regular files.

--
Stephane