Re: rm -f semantics

[Peter Eisentraut]

Gary V. Vaughan writes:

> On Sunday 11 March 2001  5:29 pm, Peter Eisentraut wrote:
> > 2001-03-11  Peter Eisentraut  <address@hidden>
> >
> >     * ltmain.in (clean,uninstall): Do not error if the file doesn't
> >     exist and 'rm -f' was used.  Exit with status 1 on error.
>
> Thanks.  Applied.

Oops, I used an (unportable) test -e in there.  I have this as an
alternative:

diff -u -r1.200.2.58 ltmain.in
--- ltmain.in   2001/04/07 00:32:39     1.200.2.58
+++ ltmain.in   2001/04/07 20:41:12
@@ -5104,7 +5104,13 @@
       fi

       # Don't error if the file doesn't exist and rm -f was used.
-      if test ! -e "$file" && test "$rmforce" = yes; then
+      # Does not support device, pipe, or socket files.
+      if test -L "$file" || test -f "$file"; then
+        :
+      elif test -d "$file"; then
+        exit_status=1
+        continue
+      elif test "$rmforce" = yes; then
         continue
       fi
===snip

This also handles symlinks correctly, which the existing one doesn't.
(Note that 'rm' removes the symlink, not the underlying file, but any
'test' other than '-L' follows the symlink.)  The problem is obviously
that test -L is most likely not portable either (especially to systems
that don't have symlinks), but that can be tested for.  How about
something along the lines of

if test $LN_S != 'ln -s'; then
  test_L='test -L'
else
  test_L=false
endif

The other issue is whether or not it seems acceptable to not support the
deletion of special files through libtool.  It seems unlikely that a build
system would deal with these in the first place.

Comments?

-- 
Peter Eisentraut      address@hidden       http://yi.org/peter-e/