find-ing / glob-ing for date(1) format

halhen · 2010-10-11 18:14:01

Is there a nice way to find filesystem objects by date(1)/strftime format?

In my backup script, I'm keeping a few old copies of each backup. Still, I'd like to configure the destination directory for each backup like e.g.

/backup/desktop-%Y%m%d

To find which existing backups there are, I need to search the file system for matching directory names. I haven't figured out a good way to do so, so I've settled with the following hackish script. It feels like there should be a better way, though. Is there?

#!/bin/bash
[[ -z $1 ]] && echo "Usage: $(basename $0) <pattern>" && exit 1

# Convert strftime sequences to wildcards
dir_glob=$(echo $1 | sed 's/%./*/g;s/*\+/*/g')

# First, match by wildcards
for cand in $(ls -d1 $dir_glob 2>/dev/null); do
    # Second, make sure the wildcards were valid FORMATS
    python -c "import time, sys; time.strptime('$cand', '$1'); sys.exit(0);" 2>/dev/null
    [[ $? -eq 0 ]] && echo $cand
done

Example use:

find_by_strftime_format /backup/desktop-%Y%m%d

Last edited by halhen (2010-10-11 18:16:37)

karol · 2010-10-11 21:40:03

You can use 'find / -type d -name <pattern>' or 'find / -type d -regex <pattern>' to look for some directories. Maybe you can use '-exec' instead of the loop.

The <pattern> 'date +%Y%m%d' is just 8 digits, with '20' being the first two, right?

Last edited by karol (2010-10-11 22:01:09)

halhen · 2010-10-11 22:17:47

karol wrote:

The <pattern> 'date +%Y%m%d' is just 8 digits, with '20' being the first two, right?

Yeah, that's right. I don't want to enter a second configuration stating that %Y%m%d is eight digits. First and foremost since I hate unnescessary duplication and it should be possible to change formats without having to figure out a regular expression to go with it, but also since there are more rules than eight digits to the pattern. 20101355 is for example not a valid date.

karol · 2010-10-11 22:19:28

If you're searching for a directory named e.g. 'desktop-20100603', you can use

find / -type d -regextype posix-extended -regex ".*desktop-2010(0[1-9]|10|11|12)(0[1-9]|(1|2)[0-9]|30|31)" \! -regex ".*desktop-2010(0(229|230|231|431|631|931)|1131)"

The search was I/O bound and took over a minute here. YMMV of course :-)

Edit: I hope the validation is OK now ;P

Last edited by karol (2010-10-11 23:15:54)

Arch Linux

#1 2010-10-11 18:14:01

find-ing / glob-ing for date(1) format

#2 2010-10-11 21:40:03

Re: find-ing / glob-ing for date(1) format

#3 2010-10-11 22:17:47

Re: find-ing / glob-ing for date(1) format

#4 2010-10-11 22:19:28

Re: find-ing / glob-ing for date(1) format

Board footer