[pacman-dev] [arch-general] rankmirrors with arch-games

[Dan McGee]

On Thu, Mar 18, 2010 at 9:42 PM, Daenyth Blank <daenyth+arch at gmail.com> wrote:
> For some reason rankmirrors seems to really hate the arch-games
> mirrorlist. I don't know whether it's a bug or whether there's some
> sort of server misconfiguration.
>
> [root at Muspelheimr pacman.d]# rankmirrors -t archgames-mirrorlist
> Querying servers, this may take some time...
>  *   *   *   *
>  Servers sorted by time (seconds):
> http://pseudoform.org/arch-games/games/i686 #(Nürnberg, Germany.
> Maintainer: svenstaro) : 1.01
> http://repo.exigen.org/arch/games/i686 #(Düsseldorf, Germany.
> Maintainer: s4msung) : unreachable
> http://repo.archlinux-gaming.org/i686 #(Toronto, Canada. Primary
> mirror) : unreachable
> ftp://mirror.selfnet.de/arch-games/i686 #(Stuttgart, Germany.
> Maintainer: hrist) : unreachable
>
> All of the above servers respond to ping and their repos can be
> browsed manually and synced with using pacman.

So let's compare a "normal" mirror location with those from arch-games
and what we know works.

normal: ftp://ftp.archlinux.org/core/os/i686
works: http://pseudoform.org/arch-games/games/i686
not work: http://repo.exigen.org/arch/games/i686

rankmirrors assumes the repository name is the third component from
the end; the normal URL would thus produce "core" and look for
"core.db.tar.gz" and the working arch-games URL would see "arch-games"
and look for "arch-games.db.tar.gz". The non-working URLs thus are
guessing the db name incorrectly; the one I listed above would look
for "arch.db.tar.gz" and fail to find it, thus the unreachable in the
output.

I cross-posted this to pacman-dev, but we have a bit of a oddity here
as the rankmirrors python script was recently replaced by a bash
version, and it does an absolutely abominable job on this mirrorlist
due to the inline comments:

$ rankmirrors test-mirrorlist
# Server list generated by rankmirrors on 2010-03-20
#
# Arch Games repository mirrorlist
#
url `' is malformed.

I'm not super thrilled about this regression. Either way, I think we
should probably add an option to the scripts to use a designated file
as the target for rankmirrors testing; this way you could specify a DB
filename or any other file as the target to test against.

-Dan