On 29/09/10 22:41, Allan McRae wrote:

On 29/09/10 22:33, Allan McRae wrote:

...

This prevents circular symlinks and weird final package locations when using commands like 'PKGDEST="." makepkg'.

Fixes FS#20922.

Signed-off-by: Allan McRae<allan@archlinux.org> ---

Note that this adds readlink as a dependency to makepkg. That is provided by coreutils which is essential anyway so I do not think this is an issue.

And I may be wrong... it appears readlink does different things with -f on OSX/BSD. Off to find a more compatible solution.

Could any BSD/OSX users report on whether "realpath ." give the full path? Thanks, Allan

On 29/09/10 23:25, Dan McGee wrote:

On Wed, Sep 29, 2010 at 8:15 AM, Allan McRae<allan@archlinux.org> wrote:

...

On 29/09/10 22:41, Allan McRae wrote:

...

On 29/09/10 22:33, Allan McRae wrote:

...

This prevents circular symlinks and weird final package locations when using commands like 'PKGDEST="." makepkg'.

Fixes FS#20922.

Signed-off-by: Allan McRae<allan@archlinux.org> ---

Note that this adds readlink as a dependency to makepkg. That is provided by coreutils which is essential anyway so I do not think this is an issue.

And I may be wrong... it appears readlink does different things with -f on OSX/BSD. Off to find a more compatible solution.

Could any BSD/OSX users report on whether "realpath ." give the full path?

It worked fine according to this: c8a41b7d6da7f820754a07cb085687ea5e110f85

Just to clarify (thanks to Xavier), it worked on BSD but not in OSX which does not have either "readlink -f" or "realpath"... Argh!

On 29/09/10 23:58, Xavier Chantry wrote:

On Wed, Sep 29, 2010 at 3:52 PM, Allan McRae<allan@archlinux.org> wrote:

...

Just to clarify (thanks to Xavier), it worked on BSD but not in OSX which does not have either "readlink -f" or "realpath"...

just found a link that might help : http://stackoverflow.com/questions/1055671/how-can-i-get-the-behavior-of-gnu...

To sum up the solutions : 1) use python os.path.realpath 2) use C realpath (3) 3) use fink /sw/sbin/readlink -f 4) implement in bash

Someone already proposed 4) a while ago : http://mailman.archlinux.org/pipermail/pacman-dev/2009-February/008147.html I did not compare this bash implementation with the one found on stackoverflow.

The best might still be to see if there is any way to avoid using realpath, like I did for repo-add back then. But maybe there is really no way out for makepkg, I don't know.

Well, one way to avoid all this is to add something like: [[ $PKGDIR == "." ]] && $PKGDIR="$startdir" before we preserve the environmental variables. But that is bad as things like PKGDIR="../" would still cause issues. So I think we should just add the bash implementation of realpath as a function inside makepkg. The one at stackoverflow is very simplistic so I will take another look at the one posted to the mailing list. We should really look at libifying some/all of this sort of stuff out of makepkg... Allan

On 30/09/10 21:26, Xavier wrote:

...

Well, one way to avoid all this is to add something like:

[[ $PKGDIR == "." ]]&& $PKGDIR="$startdir"

before we preserve the environmental variables. But that is bad as things like PKGDIR="../" would still cause issues.

I rather meant a way to remove any assumption of PKGDEST being absolute.

This is the only broken code, isn't it :

if (( ! ret ))&& [[ "$PKGDEST" != "${startdir}" ]]; then ln -sf "${pkg_file}" "${pkg_file/$PKGDEST/$startdir}" ret=$? fi

No. It also affects the location of the tarfile. e.g. PKGDEST="." puts the tar file in $pkgdir because we create the tarball location using $PKGDEST while in $pkgdir: cd $pkgdir ... pkg_file="$PKGDEST/${nameofpkg}-${pkgver}-${pkgrel}-${PKGARCH}${PKGEXT}" ...

Why not do something like : local link_file="$startdir/${nameofpkg}-${pkgver}-${pkgrel}-${PKGARCH}${PKGEXT}" if [[ ! -f "$link_file" ]]; then ln -sf $pkg_file $link_file fi

I don't know if there are cases where a regular file exists but is not the package we just built. If it can happen and we care, we could try to test whether pkg_file and link_file are not actually the same regular file.

Hmm... we check if the package exist before building but do we check both PKGDEST and $startdir? We probably should.

...

So I think we should just add the bash implementation of realpath as a function inside makepkg. The one at stackoverflow is very simplistic so I will take another look at the one posted to the mailing list. We should really look at libifying some/all of this sort of stuff out of makepkg...

Btw there are actually several bash impl in that link, the last one is more complex : http://stackoverflow.com/questions/1055671/how-can-i-get-the-behavior-of-gnu...

Ah - I had missed that. That is definitely the best implementation I have seen so far. Although the dereferencing the symlink is simpler / more efficient in the one posted on this mailing list. Some combination might be fine. Allan

On 01/10/10 01:29, Xavier wrote:

Allan McRae wrote:

...

No. It also affects the location of the tarfile. e.g. PKGDEST="." puts the tar file in $pkgdir because we create the tarball location using $PKGDEST while in $pkgdir:

cd $pkgdir ... pkg_file="$PKGDEST/${nameofpkg}-${pkgver}-${pkgrel}-${PKGARCH}${PKGEXT}" ...

ok but do we need to be in $pkgdir ? IIRC this is exactly what I had to change in repo-add (the current working dir). The problem is always the same : if we change dir, then relative paths become broken. Looking at the create_package function, it seems we just need to add a few $pkgdir/ here and there.

I was looking at the makeworld script in ABS and noticed it has a nice way of dealing with this:

PKGDEST="." cd $PKGDEST PKGDEST=$(pwd) cd - &>/dev/null echo $PKGDEST /home/allan

Now that is what I call a simple solution to implementing readlink/realpath in bash! I'll adjust my patch. Allan