bash script regex to get directory path up nth levels

I'm using PWD to g开发者_运维技巧et the present working directory. Is there a SED or regex that I can use to, say, get the full path two parents up?

Why sed or regex? Why not dirname:

parent=`dirname $PWD`
grandparent=`dirname $parent`

Edit:

@Daentyh points out in the comments that apparently $() is preferred over backquotes `` for command substitution in POSIX-compliant shells. I don't have experience with them. More info:

http://www.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html#tag_02_06_03

So, if this applies to your shell, you can (should?) use:

parent=$(dirname $PWD)
grandparent=$(dirname $parent)

This should work in POSIX shells:

echo ${PWD%/*/*}

which will give you an absolute path rather than a relative one.

Also, see my answer here where I give two functions:

cdn () { pushd .; for ((i=1; i<=$1; i++)); do cd ..; done; pwd; }

which goes up n levels given n as an argument.

And:

cdu () { cd "${PWD%/$1/*}/$1"; }

which goes up to a named subdirectory above the current working directory.

why not use

"${PWD}/../.."

?

Not sed or regex, but this does do arbitrary parent quantity:

$ cd $(mktemp -dp $(mktemp -dp $(mktemp -dp $(mktemp -d)))) && \
> n=3 && \
> readlink -f  ${PWD}/$(for i in $(seq ${n}); do echo -n '../' ; done)
/tmp/tmp.epGcUeLV9q

In this example, I cd into a 5-deep temporary directory, assign n=3, construct a relative path n levels up from ${PWD}, and -f, --canonicalize the result with readlink.

Here's the regex that works for me. It's a little different between grep and Perl/sed:

The extended regex breaks up paths into 0 or more groups of /ABC123 anchored to the end of line, essentially working backward. (.*) consumes everything prior to this match and replaces it.

user@home:~/adir/bdir$ pwd
/home/user/adir/bdir

user@home:~/adir/bdir$ pwd | perl -pe 's|(.*)((/.*?){0})$|\1|'
/home/user/adir/bdir

user@home:~/adir/bdir$ pwd | perl -pe 's|(.*)((/.*?){1})$|\1|'
/home/user/adir

user@home:~/adir/bdir$ pwd | sed -r 's|(.*)((/.*?){2})$|\1|'
/home/user

user@home:~/adir/bdir$ pwd | sed -r 's|(.*)((/.*?){3})$|\1|'
/home

user@home:~/adir/bdir$ pwd | sed -r 's|(.*)((/.*?){4})$|\1|'

Grep can simulate substitution using a positive look ahead (?= which tells grep to match everything except the pattern. -Po tells grep to use Perl regex and show only the match.

user@home:~/adir/bdir$ pwd
/home/user/adir/bdir

user@home:~/adir/bdir$ pwd | grep -Po '(.*)(?=((/.*?){0})$)'
/home/user/adir/bdir

user@home:~/adir/bdir$ pwd | grep -Po '(.*)(?=((/.*?){1})$)'
/home/user/adir

user@home:~/adir/bdir$ pwd | grep -Po '(.*)(?=((/.*?){2})$)'
/home/user

user@home:~/adir/bdir$ pwd | grep -Po '(.*)(?=((/.*?){3})$)'
/home

user@home:~/adir/bdir$ pwd | grep -Po '(.*)(?=((/.*?){4})$)'

Of course it works equally well for Windows style paths:

C:\home\user\adir\bdir> cd
C:\home\user\adir\bdir

C:\home\user\adir\bdir> cd | perl -pe 's|(.*)((\\.*?){0})$|\1|'
C:\home\user\adir\bdir

C:\home\user\adir\bdir> cd | sed -r 's|(.*)((\\.*?){1})$|\1|'
C:\home\user\adir

C:\home\user\adir\bdir> cd | grep -Po '(.*)(?=((\\.*?){2})$)'
C:\home\user

C:\home\user\adir\bdir> cd | grep -Po '(.*)(?=((\\.*?){3})$)'
C:\home

C:\home\user\adir\bdir> cd | grep -Po '(.*)(?=((\\.*?){4})$)'
C:

Sorry for the edits but I've been working on this enigma for about 16 hours now. Just kept trying different permutations and re-reading the regex docs. It had to sink in eventually.