On Sat, Nov 24, 2007 at 02:24:02PM +0100, Nagy Gabor wrote:

...

I didn't understand why realpath was called on every files of every filelist in query_fileowner : ppath = resolve_path(path);

It turns out this is needed for the diverted files. For example, cddb_get installs /usr/lib/perl5/site_perl/5.8.8/CDDB_get.pm which actually ends in /usr/lib/perl5/site_perl/current/CDDB_get.pm .

A real life example ;-): /opt directory can be a symlink for example.

And an offtopic note here: iirc (<- cannot check now) you cannot search for owner of a directory symlink [/usr/var], because dir-symlinks are treated as dirs in files <- this may(?) lead to an other symlink puzzle: conflicting "symlinks" in target list. Xavier, could you check this please?

Not sure what you mean, but usr/var is owned by filesystem and show up as usr/var in the filelist. So it's treated as a file and not a dir, right? pacman -Ql filesystem | grep "usr/var" filesystem /usr/var

On Sat, Nov 24, 2007 at 02:59:05PM +0100, Nagy Gabor wrote:

...

Not sure what you mean, but usr/var is owned by filesystem and show up as usr/var in the filelist. So it's treated as a file and not a dir, right? pacman -Ql filesystem | grep "usr/var" filesystem /usr/var

[*] Fine. First I meant "pacman -Qo /usr/var", when I tried last time, "I got cannot determine ownership of a directory" <- if this still exists, we can fix with the help of [*].

Hm, I don't understand your [*] thing. But pacman -Qo simply stats the file (so if it's a symlink, it resolves it), and if it is a directory, it simply stops. That's because directories are usually owned by multiple packages, and -Qo is maybe supposed to return just one package.

On Nov 26, 2007 4:09 PM, Aaron Griffin wrote:

On Nov 25, 2007 2:24 PM, Dan McGee wrote:

...

I feel like their could be an edge case where something breaks and this shortcut bites us. However, I can't figure out out. Anyone else? Otherwise this makes a lot of sense.

I don't think so.

operation 1 : check basename if same operation 2 : check full path if same tada!

The only edge cases are when basename matches, but the full path does not, which is covered anyway.

Personally, I think this is a brilliant idea on Xavier's part, so keep up the good work.

Also, make sure to note what Dan mentioned about basename (we want to try and be cross-platform here whenever possible)

You're a day behind. :P http://projects.archlinux.org/git/?p=pacman.git;a=commit;h=11133da587ebc1c78... http://projects.archlinux.org/git/?p=pacman.git;a=commit;h=d683033d3ea79956f... -Dan

On Mon, Nov 26, 2007 at 03:09:44PM -0600, Aaron Griffin wrote:

On Nov 25, 2007 2:24 PM, Dan McGee wrote:

...

I feel like their could be an edge case where something breaks and this shortcut bites us. However, I can't figure out out. Anyone else? Otherwise this makes a lot of sense.

I don't think so.

operation 1 : check basename if same operation 2 : check full path if same tada!

The only edge cases are when basename matches, but the full path does not, which is covered anyway.

Personally, I think this is a brilliant idea on Xavier's part, so keep up the good work.

Also, make sure to note what Dan mentioned about basename (we want to try and be cross-platform here whenever possible)

Oh, sorry, we discussed this with Dan on IRC, and then I was willing to answer the mail so that others can follow, and totally forgot. When Dan asked me about these edge cases, I wasn't sure what to answer. But when thinking back about it later, I thought about something similar to what you just did for proving it was correct, so thanks :) About the basename, I did what Dan asked, move mbasename from pacman.c to util.c , and then use that instead. That change and the Qo patch are both already in master.