{"id":474,"date":"2008-10-30T17:20:49","date_gmt":"2008-10-31T00:20:49","guid":{"rendered":"http:\/\/www.stevenbrown.ca\/blog\/?p=474"},"modified":"2008-11-18T12:23:31","modified_gmt":"2008-11-18T20:23:31","slug":"open-with-for-the-command-line","status":"publish","type":"post","link":"https:\/\/www.stevenbrown.ca\/blog\/archives\/474","title":{"rendered":"open-with for the command-line"},"content":{"rendered":"

Update 2008\/11\/18<\/strong>: Use xargs<\/code> \ud83d\ude1b<\/p>\n

Here’s a bash script that you can pipe output into and tell it to run a specific program with the output as arguments. I’ve named it open-with<\/code> and placed it in my personal script directory: \/home\/steve\/bin\/<\/code>. Look within the script at a couple of the examples for how to use it.<\/p>\n

\r\n#!\/bin\/bash\r\n\r\nPROG=`basename $0`\r\n\r\nDESC=\"Read arguments from standard input and run a specified program\r\nwith them.  Meant to be used as output for a pipe.\"\r\n\r\n\r\nUSAGE=\"OUTPUT | $PROG \\\"PROGRAM-NAME [OPTIONS AND ARGUMENTS]\\\"\"\r\n\r\nEXAMPLES=\"\r\n# List (with details), all the files that include \\\"downloads\\\" in their name:\r\n  find \/home\/ -iname \\\"*downloads*\\\" | $PROG \\\"ls -l\\\"\r\n\r\n# Queue all AVI files in current directory in vlc:\r\n  ls *.avi | $PROG vlc\r\n\r\n# View all time-related icons with eye-of-gnome:\r\n  find \/usr\/share\/icons\/gnome\/ -iregex \\\".*[^mul]time.*\\\" | open-with eog\r\n\"\r\n\r\nfunction Usage() {\r\n  echo \"DESCRIPTION: $DESC\" ; echo\r\n  echo \"USAGE: $USAGE\" ; echo\r\n  echo -n \"EXAMPLES: $EXAMPLES\"\r\n}\r\n\r\n# Quick check to see it's being called correctly, if not, print Usage and exit\r\nif [ ${#@} -ne 1 ]\r\nthen\r\n\tUsage\r\n\texit\r\nfi\r\n\r\nfiles=()\t\t\t\t\t\t# empty array\r\nwhile read -r\t\t\t\t\t# read from stdin\r\ndo\r\n\tfiles+=( \"$REPLY\" )\t\t\t# add result of read to array\r\ndone\r\n\r\n# assume $1 is a valid program\r\n$1 \"${files[@]}\"\t\t\t\t# pass arguments to specified program\r\n<\/pre>\n
Motivation<\/h3>\n
You’re sitting at the command line and have a list of images (in different locations) that you would like to browse.  Most image viewers let you iterate over a set of images, but only within the same directory.  What would be great is if they accepted input from stdin through a pipe!<\/p>\n
\r\n$ cat my_list_of_images | my_image_viewer\r\n<\/pre>\n
I didn’t find anything that did that, but using the open-with<\/code> script, you can do something similar:  <\/p>\n
\r\n$ cat my_list_of_images | open-with my_image_viewer\r\n<\/pre>\n
\"\"<\/a>
\nBrowsing a select list of images is actually kind of nice.  It’s like a playlist for your image viewer – a viewlist.  \ud83d\ude42  Anyway, I’m sure there’s some problems with this script.  Feel free to provide suggestions in the comments.  But I certainly don’t want to look at it for a while….<\/p>\n
I hate shell scripting<\/h3>\n
With a passion.  It’s not a great surprise… many programmers do.  I’m willing to go on record and state that I hate it even more than PERL programming.  I rarely do it, and when I decide to try something that seems like it would be simple, it turns out taking forever due mostly to quirks.  The rest of this post is a bit about how I went about writing this script, which ended up being mostly given to me by some #bash<\/code> gurus on IRC.  And it’s a bit of a rant.<\/p>\n
Although not many GUI programs seem to accept stdin as input, most accept filenames as arguments:<\/p>\n
\r\n$ eog image1.jpg image2.jpg image3.jpg \"\/home\/steve\/image seven.jpg\"\r\n<\/pre>\n
So I figured I would just have to convert the list of images into an acceptable format: quoted, absolute filenames, separated by space.  Doing this depends entirely on the format the list is currently in, but it’s likely a list of unquoted filenames separated by newlines:<\/p>\n
\r\n\/home\/steve\/image1.jpg\r\n\/home\/steve\/my images\/wow.jpg\r\n\/tmp\/anotherimage.png\r\n<\/pre>\n
For me, my test list looked like this:<\/p>\n
\r\n\/usr\/share\/icons\/gnome\/48x48\/stock\/generic\/stock_timezone.png\r\n\/usr\/share\/icons\/gnome\/16x16\/stock\/generic\/stock_timezone.png\r\n\/usr\/share\/icons\/gnome\/16x16\/stock\/generic\/stock_timer.png\r\n\/usr\/share\/icons\/gnome\/16x16\/stock\/generic\/stock_timer_stopped.png\r\n\/usr\/share\/icons\/gnome\/16x16\/stock\/form\/stock_form-time-field.png\r\n\/usr\/share\/icons\/gnome\/24x24\/stock\/generic\/stock_timezone.png\r\n\/usr\/share\/icons\/gnome\/24x24\/stock\/generic\/stock_timer.png\r\n\/usr\/share\/icons\/gnome\/24x24\/stock\/generic\/stock_timer_stopped.png\r\n\/usr\/share\/icons\/gnome\/24x24\/stock\/form\/stock_form-time-field.png\r\n<\/pre>\n
I was looking for icons of clocks<\/em> or representations of time<\/em>, and I obtained that list using find<\/code>:<\/p>\n
\r\n$ find \/usr\/share\/icons\/gnome\/ -iregex \".*[^mul]time.*\"\r\n<\/pre>\n
So I can’t pipe it into my image viewer, but I can use the output as the command-line arguments if I change the newlines to spaces.  find<\/code> has an option for formatting the output which is perfect:<\/p>\n
\r\n$ find \/usr\/share\/icons\/gnome\/ -iregex \".*[^mul]time.*\" -printf \"'%p' \"\r\n'\/usr\/share\/icons\/gnome\/48x48\/stock\/generic\/stock_timezone.png' '\/usr\/share\/icons\/gnome\/16x16\/stock\/generic\/stock_timezone.png' '\/usr\/share\/icons\/gnome\/16x16\/stock\/generic\/stock_timer.png' '\/usr\/share\/icons\/gnome\/16x16\/stock\/generic\/stock_timer_stopped.png' '\/usr\/share\/icons\/gnome\/16x16\/stock\/form\/stock_form-time-field.png' '\/usr\/share\/icons\/gnome\/24x24\/stock\/generic\/stock_timezone.png' '\/usr\/share\/icons\/gnome\/24x24\/stock\/generic\/stock_timer.png' '\/usr\/share\/icons\/gnome\/24x24\/stock\/generic\/stock_timer_stopped.png' '\/usr\/share\/icons\/gnome\/24x24\/stock\/form\/stock_form-time-field.png' \r\n<\/pre>\n
That’s great for me, because I’m using find<\/code><\/em>.  But not very useful if I’m not, so I wanted something more generic.  Of course, there are many ways to do this, and again, I’m by no means a command-line guru.  But here’s how I started:<\/p>\n
\r\n$ find \/usr\/share\/icons\/gnome\/ -iregex \".*[^mul]time.*\" | sed -e 's\/^\/\"\/' | sed -e :a -e '$!N;s\/\\n\/\" \/; ta' | tr '\\n' '\"'\r\n<\/pre>\n
This wrapped the lines in double-quotes and join them together with a space in between.  The find<\/code> command is the same as before, minus the formatting because that’s what I was trying to find an alternative to.  The output of find<\/code> is piped into sed<\/code>, which adds a \"<\/code> at the beginning of each line.  This output is then sent to another sed<\/code> which replaces the newline character at the end of each line with a closing double-quote and a space, joining all lines into a single line.  Finally, that output is piped into tr<\/code> which replaces the one remaining newline with a final double-quote.<\/p>\n
If the files don’t include spaces or other troublesome characters (mine didn’t), then you could get away with simply changing the newlines into spaces.  But again, I wanted something generic.<\/p>\n
\r\n$ find \/usr\/share\/icons\/gnome\/ -iregex \".*[^mul]time.*\" | tr '\\n' ' '\r\n<\/pre>\n
Anyway, now that we have something that creates the desired input, we just have to wrap it in back-ticks and put the whole mess as the argument to the image viewer!  In my case, I’m using eye-of-gnome (or eog).<\/p>\n
\r\n$ eog `find \/usr\/share\/icons\/gnome\/ -iregex \".*[^mul]time.*\" | sed -e 's\/^\/\"\/' | sed -e :a -e '$!N;s\/\\n\/\" \/; ta' | tr '\\n' '\"'`\r\n<\/pre>\n
Wait a second.  That doesn’t actually work.  Why not?  Running the backtick contents by itself seemed to produce the correct output.  Copying this output verbatim as arguments to eog<\/code> worked as expected.  The problem was that when the quoted arguments were manually entered on the command-line, bash (silently) escapes the filenames, removing the quotes, adding backslashes before space,s etc.  But when the pipeline is wrapped in backticks, the result is not escaped, and eog<\/code> complains about not being able to find files that begin with quotes.  Fine.  I was not about to write a bash “escape” script – something like that should already exist, right?  And it should be built in to bash!  Perhaps it’s even called “escape<\/code>“.  Well, if there exists such a built-in, I couldn’t find it.  But I had to be going about this the wrong way.  Heading over to IRC, it was kind of difficult to explain the problem, but ferret<\/code> eventually gave me the meat of the script above:<\/p>\n
\r\nfiles=(); while read -r; do files+=( \"$REPLY\" ); done < <(find \/usr\/share\/icons\/gnome\/ -iregex \".*[^mul]time.*\"); eog \"${files[@]}\"\r\n<\/pre>\n
It worked!  I thanked him and began studying it.  There were still a couple things I didn't understand:<\/p>\n