Follies with find

[curmudgeon]

I am trying to use find to do something where I need each file as an argument twice.

For an example, let's suppose I wanted to set all of the access times in a tree equal to the modification time.

I thought something like this would work (but it doesn't):

Code: Select all

$ find /usr/local/portage/sys-kernel/gentoo-sources/ -exec touch -d "`date -r {}`" {} ';'

Sure, I could do this (fairly easily even) in two steps - do the find, listing the files and the dates (saving the output to a file), and then editing the file (turning it into a script by putting the touch command at the front of each line), but I am looking for an elegant (meaning one command) solution.

Any ideos?

[jongeek]

Ok, so I don't know how to do it with one command, but you can use bash to do it in one line.

Code: Select all

for file in $(find /usr/local/portage/sys-kernel/gentoo-sources); do touch -d "$(date -r ${file})" ${file}; done

I hope that follows the spirit of your request, if not the letter.

[aderesch]

I am trying to use find to do something where I need each file as an argument twice.

Code: Select all

$ find /usr/local/portage/sys-kernel/gentoo-sources/ -exec touch -d "`date -r {}`" {} ';'

Your problem isn't using an argument twice, but rather the backticks (command substitution).
This should do, although it's not elegant:

Code: Select all

find -exec bash -c 'touch -d "`date -r "{}"`" "{}"' \;

ad

Edit: Plus it is of course inefficient, but that's not what you asked. :-) So in reality I would also prefer the above for-loop approach.
Edit2: Changed to use single quotes instead of lots of escaping.

[Akkara]

A idiom I like to use for such things goes like this:

Code: Select all

fine $DIR ...conditions... -print | while read F; do
    echo "$F" "and again" "$F"
done

This works even for files and paths with spaces and special characters in the name (but not if there's a newline in the name).

[mv]

And the right way to do things like this (which works however the filename looks like) is the following:

Code: Select all

find /usr/local/portage/sys-kernel/gentoo-sources/ -exec /bin/sh -c 'for i
do	touch -d "`date -r -- "${i}"`" -- "${i}"
done' sh '{}'  '+'

Of course, changing the date of the file to its own date makes no change, but I suppose this was just an example (in connection with "touch" you will problably want to use the find option -depth anyway).

[curmudgeon]

Of course, changing the date of the file to its own date makes no change, but I suppose this was just an example (in connection with "touch" you will problably want to use the find option -depth anyway).

It's supposed to, but it doesn't seem to work.

From the man page:

Update the access and modification times...

But:

Code: Select all

# ls -cl gentoo-sources-2.6.32-r7.ebuild
-rw-r--r-- 1 portage portage 899 2010-06-13 20:22:24 gentoo-sources-2.6.32-r7.ebuild

# touch -d 2010-04-10 gentoo-sources-2.6.32-r7.ebuild

# ls -cl gentoo-sources-2.6.32-r7.ebuild
-rw-r--r-- 1 portage portage 899 2010-06-13 20:23:08 gentoo-sources-2.6.32-r7.ebuild

I get similar results trying touch -a -d.

[mv]

ls -cl shows the ctime; this is neither the atime nor the mtime.

[Hu]

(but not if there's a newline in the name).

For that, use -print0 and read -d $'\0' to generate and read null terminated records. This requires GNU find and bash, rather than just find and POSIX sh as yours does, but should get the job done if those requirements are satisfied. Unlike the construct proposed by mv, this lets you retain a find/while read pattern, thus avoiding the creation of one shell per record.

[mv]

to generate and read null terminated records. [...]
Unlike the construct proposed by mv, this lets you retain a find/while read pattern, thus avoiding the creation of one shell per record.

In the first sentence you seem to mean by "record" 1 filename. However, in the solution which I posted, one record is typically several thousand filenames (determined by the maximal length of a command). Moreover, you save some time since you do not need to send the data over a pipe and can use dash (much faster than bash, especially when spawning); which solution is actually faster altogether depends on a lot of unknown data. So if in doubt, I would prefer the compatible solution over the GNUism.

[Hu]

However, in the solution which I posted, one record is typically several thousand filenames (determined by the maximal length of a command). Moreover, you save some time since you do not need to send the data over a pipe and can use dash (much faster than bash, especially when spawning); which solution is actually faster altogether depends on a lot of unknown data. So if in doubt, I would prefer the compatible solution over the GNUism.

Indeed, I overlooked your use of the SVR4 variation. Too often, I see people use the traditional model of spawning one process per filename.

[curmudgeon]

ls -cl shows the ctime; this is neither the atime nor the mtime.

Oops. Guess I didn't understand that too well (just did a bunch more reading on the subject).

It seems this would be the tool to manipulate timestamps:

http://stroke.sourceforge.net/