Re: [PATCH] _git: offer changed files relative to current directory

["Alexey I. Froloff" <raorn@altlinux.org>]

[-- Attachment #1: Type: text/plain, Size: 2044 bytes --]

On Sat, Dec 05, 2009 at 06:00:53PM +0100, ??t??p??n N??mec wrote:
> I'm not sure about the exact purpose of this function, but it doesn't
> seem correct to me:
Yes, my code requires some explanations.

$what is the file we got from "git tree-index" or "git ls-files
--full-name -- $(git rev-parse --show-cdup)", we can assume it's
a full path from the root of repository.

$to is the path we got from "git rev-parse --show-prefix".

    what=( ${(s:/:)"${${${${1//\/\///}//\/.\///}%%/.}%%/}"} )
    to=( ${(s:/:)"${${${${2//\/\///}//\/.\///}%%/.}%%/}"} )

Some paranoid code that strips "/./", converts multiple slashes
to one and strips "/." and "/" from right.  Then path is split to
components for easier processing.

BTW, if prefix is empty string, $to will be equal to ( "" ) which
will add unneeded "../" later (my bug).

    while (( $#what > 0 )) && (( $#to > 0 )) && [[ $what[1] == $to[1] ]]; do
	what[1]=()
	to[1]=()
    done

This loop strips common prefix from $what and $to, we wont need
it in relative path.

    while (( $#to > 0 )); do
	res+=..
	to[1]=()
    done

For each path component left in $to ascend one level up.

    res=( "$res[@]" "$what[@]" )

Add path components left in $what.

> It is not unlikely to have more same-named directories in different
> parts of directory tree. IIUC, the code above won't always work
> correctly in such situation (i.e. it will remove the directory from the
> path, wrongly assuming it is the same directory for both paths).
No, first loop will only remove prefix common for both paths.

> I believe the only correct way to solve the path relativization problem
> for arbitrary input is to use the *absolute* paths.
These paths _are_ absolute since it is impossible to ascend
outside of git repository.  Hope we can assume so ;-)

> In any case, it would be more useful/comprehensible (to me at least) if
> you included the solution with the former patch to solve the actual
> problem.
A bit later.

-- 
Regards,
Sir Raorn.

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]