Arch Linux

$ makepkg -sr
==> Making package: mail-notification 5.0-1 (середа, 30 січня 2008 17:38:51 +0200)
==> Checking Runtime Dependencies...
==> Installing missing dependencies...
resolving dependencies...
looking for inter-conflicts...

Targets: postgresql-libs-8.2.6-1 libmysqlclient-5.0.51-1
cyrus-sasl-plugins-2.1.22-6 sharutils-4.6.3-1 gmime-2.2.15-1

Total Download Size: 0,36 MB
Total Installed Size: 3,71 MB

Proceed with installation? [Y/n]
:: Retrieving packages from extra...
gmime-2.2.15-1-i686 364,2K 6,3K/s 00:00:58 [#####################] 100%
checking package integrity...
(5/5) checking for file conflicts [#####################] 100%
(1/5) installing postgresql-libs [#####################] 100%
(2/5) installing libmysqlclient [#####################] 100%
(3/5) installing cyrus-sasl-plugins [#####################] 100%
(4/5) installing sharutils [#####################] 100%
(5/5) installing gmime [#####################] 100%
==> Checking Buildtime Dependencies...
==> Installing missing dependencies...
resolving dependencies...
looking for inter-conflicts...

Targets: gob2-2.0.14-1 libnetworkmanager-0.6.5-1 gtkhtml-3.16.3-1
evolution-2.12.3-1

Total Download Size: 47,62 MB
Total Installed Size: 48,65 MB

Proceed with installation? [Y/n]

==> ERROR: Aborted by user! Exiting...
==> Removing installed dependencies...
loading package data... failed.
error: failed to add target 'gob2' (could not find or read package)

It'd probably be easier for us to maintain a list of "packages installed" for the -r flag, rather than try to uninstall all everything.

Another question I had in the past that would have an impact on this :
Do we want pacman to fail when it doesn't find a target? Why not just skip it?

For removal? Skipping it makes sense to me with a warning or something

I like failing hard when something goes wrong on the command line- I think of pacman as a transactional tool in the sense that it is going to do everything I tell it, or nothing at all if something is invalid.

Hm yes, I wanted to mention your point of view here since you stated it on the ML too.
Well, I understand your position, even if I find it less practical. So forget my comment,
and this bug should be fixed the proper way like Aaron suggested.

It is almost impossible to implement this "remove dependencies" feature correctly.
Have a look at Nagy's arguments there :
http://www.nabble.com/-PATCH--No-error-in-makepkg-when-removing-deps-fails-tc16453391.html#a16493483

What is wrong with just using pacman -Qtd manually?
This seems to give a much better guarantee that the correct packages will be removed.

I agree we can do the remove dependencies properly due to provides, replaces etc.

But, to fix this specific bug, we can just not do add $deplist to $pkgdeps until the line after "if handledeps $deplist; then". Then makepkg would not add the (make)depends to the pkgdeps until they were actually installed. Patch on its way.