On Fri, 2005-08-12 at 11:35 +1200, Jonathan Purvis wrote:

A. Pagaltzis wrote:

...

That’s not very helpful to people who have no set of suitable files to check the output… can you give examples of what you expect and what you are getting?

Make yourself and empty directory and try these 3 commands:

touch file\ 1 file\ 2 file\ 3 for name in * ; do echo $name ; done for name in `ls -1 | sort -r` ; do echo $name ; done

I'd like some way of making the output of the 3rd command be the output of the 2nd, only reversed. Instead it prints a line break instead of a space as bash as split the arguments given to for on spaces.

You can get what you want by changing $IFS (bash's input field separator): IFS=; for name in `/bin/ls -1 | sort -r` ; do echo $name ; done -- Colin Palmer <colinp(a)waikato.ac.nz> University of Waikato, ITS Division

That seemed to produce the correct output, but echo was only called once. But with your idea about using IFS, i managed to get this, which does work:

IFS=" "; for name in `/bin/ls -1 | sort -r` ; do echo line: $name ; done

ls -l | sort -r | xargs echo ?

* Oliver Jones <oliver(a)deeperdesign.com> [2005-08-15 15:35]:

Or if you just want a delimited filename string then this:

[oliver] mobility:~/1$ find . -name "file*" -printf "'%f'\n" 'file 3' 'file 2' 'file 1'

Except that will fall over on filenames with spaces in them. You want to pipe `find ... -print0` to something, most of the time. (Usually xargs(1). That’s much faster than spawning one process for every single file, too, even for a pretty small number of files.) Regards, -- Aristotle Pagaltzis // <http://plasmasturm.org/>

* Oliver Jones <oliver(a)deeperdesign.com> [2005-08-15 16:10]:

...

You want to pipe `find ... -print0` to something, most of the time. (Usually xargs(1). That’s much faster than spawning one process for every single file, too, even for a pretty small number of files.)

xargs is only useful for feeding commands that accept multiple files as arguments and do the same thing to each file (eg cat).

That’s why I said “usually.” :-) Sometimes, you want to pipe the input to something else. What I was emphasizing is that find(1)’s `-print0` predicate avoids any quoting problems simply by delimiting filenames using the only character that can’t be part of a filename. You can then do anything you want with the output without running into quoting problems; xargs(1) is only one choice for processing it, but the most common because it covers many simple uses.

On a small number of files/lines xargs is likely to only execute the target command once.

Exactly, that’s why it’s useful to reduce the overhead as compared to find(1)’s `-exec` predicate, which always executes the command as many times as there are files. Regards, -- Aristotle Pagaltzis // <http://plasmasturm.org/>

* Perry Lorier <perry(a)coders.net> [2005-08-16 00:55]:

you can use "-n 1" to only do one file, and -istr to make it insert the file in instead of str, so you can do

Of course, in that case, when combining with find(1), you don’t gain anything over `find ... -exec ...`. Regards, -- Aristotle Pagaltzis // <http://plasmasturm.org/>

Advance warning: this mail ended up being pretty academical in nature… * Jonathan Purvis <jon(a)purvis.co.nz> [2005-08-12 02:50]:

But with your idea about using IFS, i managed to get this, which does work:

Filenames can legally contain newlines. (Yes, it’s rare, but it’s also possible.) The canonical way to go about this is passing `-print0` to find(1) so it terminates filenames with a NUL rather than a LF, and spliting on NULs in whatever processes the filenames next.

echo `find . -name '*E1M*' | awk '{print $2,$0}' FS='[()]' | sort | sed -e 's/[^ ]* \(.*\)/\1/' -e 's/\([ ()]\)/\\\\\1/g'`

After squinting at this command for a while, I think what you’re doing is massaging the filenames to properly sort them, then putting them back together, and finally escaping shell metachars. Is that right? This is really a task sort(1) should be up to… unfortunately it’s not, since it can’t use mulitple delimiter characters and can’t cope with NUL-separated input either (even though it can produce it, sigh). awk is useless, too. Escaping metachars wouldn’t be necessary if you kept the shell out of the picture, ie no backticks to invoke mplayer, but that’s not doable with sed(1) in the picture, since that can’t split on NULs at all… yuck. I don’t see a way to do with *well* without Perl. find . -name '*E1M*' -print0 | perl -naF'\0' -0777l0e' @k = map join( "", ( split /[()]/ )[2, 0] ), @F; print for @F[ sort { $k[$a] cmp $k[$b] } 0 .. $#k ] ' | xargs -0 mplayer Quite a mouthful. :-( This really is a task that sort(1) *should* be fully up to… splitting on NULs and being able to use multiple different as delimiter characters would make it a simple matter of saying something like find . -name '*E1M*' -print0 | sort -0z -t \( -t \) -k 3,1 | xargs -0 mplayer Alas. Regards, -- Aristotle Pagaltzis // <http://plasmasturm.org/>