I expect it's old, old news to those with more shell scripting scars: but the results of the [ -e ] test are at variance with my allegedly reasonable reading of the documentation.
For all three of sh, ksh, and the /bin/test manpages, the description of the -e test reads "file exists", unlike the other file-related tests which read "file exists and <further condition>", with <further condition> being is-writable, is-exucatable, is-readable, and the like. The manpage for /bin/test is even more emphatic in suggesting it's going to be true for a strict superset of the files for which the other tests return true - "True if file exsits (regardless of type)". However, there are arguments for which -e returns false, but a different file-related test returns true. These arguments are symlinks which don't resolve to an existing file - both symlinks that point 'nowhere', i.e. to non-existent targets (directly or indirectly), and symlinks which will error with ELOOP if stat()ed. Changing the behaviour of -e for non-resolving symlinks is almost certainly a Really Bad Idea: the existing behaviour of -e is doubtless relied on by a few million shellscripts, all more or less strongly bound to the idea that if -e returns true, there's Something There, and a strong expectation that the Something is stat()able rather than merely lstat()able. But perhaps a small change to the venerable text of the sh, ksh, and /bin/test manpages might be in order? Some form of words like "exists (target exists if a symbolic link)" might capture the actual behaviour more accurately. Yes, it's a picky point - and one I wouldn't bother raising in the Linux world, where manpages are at best impressionistic; but the pithy clarity of OpenBSD manpages is a pearl beyond price, and thus worth cleaning of even small specks. As far as testing in shell scripts whether 'things' are present - using [ -e $file -o -h $file ] catches the 'exists, maybe as a symlink which doesn't resolve' case; as could the use of stat(1) with suitable format-strings, -q, -L, and related incantatia... Cheers, Stefek
