Bug#376158: Really problematic or superfluous?
I tend to agree with Frank, unless someone can provide data on the impact on
performance.
In any case, I think this is best handled on apt's side, either by fixing the
issue or by introducing a debconf prompt. This is also not necessarily
friendly for mirror operators. If the fact that there are currently 14 pdiff
files kept on mirrors (at least ftp.d.o) for etch implies that mirrors always
keep only 2 weeks of pdiffs, this means that the bandwidth consumption for
normally-sized days (apparently 100/3 kB for etch, must be less for stable)
for the maximum number of days average update by pdiffs is under 500 kB, as
opposed to about 4 MB for a full Packages. This is under the eight of the
data transfer.
If the release-notes entry is kept, note that the text is misleading when
stating that "There has been support added to <prgn>apt</prgn> to download
only the difference between packages files."
If I wouldn't know what pdiff means, I would interpret that as meaning that
support was added for only getting .deb patches for upgrading...which would
make me way too much happy to read until I'd realize what it actually means.
The entry is also unclear on what is the problem described.
Perhaps the following note would be better (if there is really need for a
note, of course):
<sect1 id="apt-pdiff"><heading>Slower updates of APT package index
files</heading>
<p>
By default, the new <prgn>apt</prgn> uses a new way to update APT package
index files (aka "apt-get update") which should use less bandwidth and be
faster. Unfortunately, the opposite effect of making the updates slower can
happen on systems with fast network connections and doing infrequent updates.
One can force using the previous way to update by adding
<tt>Acquire::Pdiffs "false";</tt> to
<file>/etc/apt/apt.conf</file>.
</p>
</sect1>
Reply to: