NetNews Usenet Archive 1992 #30/NN_1992_30.iso/spool/comp/unix/shell/5105

Text File | 1992-12-20 | 3.3 KB | 73 lines

Newsgroups: comp.unix.shell
Path: sparky!uunet!s5!is1.is.morgan.com!is0.is.morgan.com!wpm
From: [e-mail] (W. Phillip Moore)
Subject: ksh autoload quirk
In-Reply-To: [e-mail]'s message of Wed, 16 Dec 1992 20:01:49 GMT
Message-ID: <[e-mail]>
Sender: [e-mail]
Nntp-Posting-Host: exile
Organization: Informations Services Dept., Morgan Stanley Japan
References: <[e-mail]>
Date: Fri, 18 Dec 1992 15:37:02 GMT
Lines: 59


>>>>> "David" == DWT  <[e-mail]>:

David> ksh 11/16/88d under Dell SVR4 4.0 here on ddsw1 and the ksh running
David> on a 3B2 on gagme recognize any filename in a directory in $FPATH as
David> an undefined function whether or not that name has been declared for
David> autoload with "autoload" or "typeset -fu."

I see the same behavior for the same version running on SunOS -- this
behavior is a feature/bug.

David> But there is a difference when a file in $FPATH has the same name as
David> an executable file in $PATH.  If the undefined function has been
David> formally declared with "autoload" or "typeset -fu," then ksh
David> interprets an appearance of that name on the command line as a
David> reference to the function.  If the undefined function has not been
David> declared but is recognized as such solely by dint of lying in
David> $FPATH, then an appearance of the name on the command line is taken
David> by ksh parser to mean the executable in $PATH (unless the function
David> has actually been loaded with the . command or by a call to another
David> function in the same file to whose name the file is linked).

I verified the same by tracing some scripts I wrote.  This is really
frustrating, because I would like to provide a set of utilities to our
users and make them available simple by exporting FPATH in out default
environment.  There is a noticable performance penalty if one does not
autoload, linearly dependent on the size of PATH, in addition to not being
able to provide simple wrappers to system commands.

Furthermore, subshells inherit FPATH, but do NOT inherit functions
definitions.  The shell which does all the autoloading will work as
expected, but subshells will not.

An ugly workaround is to do the following in the $ENV file:

for DIR in $( echo $FPATH | sed -e 's/:/ /g' ) ; do
  for FILE in $DIR/* ; do
    [[ -f $FILE && -x $FILE ]] && autoload ${FILE##*/}
  done
done

David> Is that the correct behavior?  Is a file name in $FPATH that hasn't
David> been declared for autoload supposed to be recognizable at all as an
David> undefined function?  I haven't seen anything about that (nor about a
David> difference in precedence from a declared undefined function) in the
David> ferkakte manual.

If by the manual you are refering to "The Kornshell" by Bolsky and Korn,
then I have also searched this in vain for documentation on the above
behavior.  I'm about ready to plow through the source, and maybe fix this.

Id there any reason NOT to simply search and autoload anything in FPATH
before earching PATH??  I assume there was a reason for coding ksh this
way.

W. Phillip Moore                                        Phone: [phone]
Information Services Department                           FAX: [phone]
Morgan Stanley Japan Ltd.  Tokyo Branch             E-mail: [e-mail]
        Ote Center Bldg. 3F, 1-1-3 Otemachi, Chiyoda-ku, Tokyo 100