Expliclty inheriting from `object` was needed in Python 2 for
new-style classes, but is no longer needed in Python 3 where all
classes implicitly inherit from `object`.

The default mode is 'rt', which leads `copyfileobj()` to failure.

The Python 3 transition introduced a regression in
2f95fde0 which assumed all of
gpg's output is UTF-8.

Closes: #980182

Signed-off-by: NNiels Thykier <niels@thykier.net>

Closes: #913965

This should also ensure that the warning is not reported by the test
suite as we want to see it.

This reverts commit 1a507077, reversing
changes made to 4b950307.

With the changes test test no longer tests the custom `__eq__` and
`__ne__` member functions, but only the result of `str`.

The DeprecationWarnings are 'comparison with a `str` is deprecated'.
Signed-off-by: NStéphane Blondon <sblondon@debian.org>

Due to a previous bug in dak, the pdiff history in Debian experimental
is currently misaligned.  Unfortunately, dak is not self-healing with
this kind of history corruption (but it had been if the field order
had been different).

This patch will have dak detect misaligned patch histories (where
e.g. `SHA1-Patches` and `SHA256-Patches` disagree on the order) and
attempt to re-align them.  We re-align by looking from the end of the
two misaligned histories and keep all the patches they have in common.

This is done on the basis that we keep adding patches at the end, so
the history should be aligned there (assumping relevant bug fixes for
the corruption source have implemented).  Once we discover a
difference, we discard that and all older patches.  This is a
deliberate choice as our PDiff code assumes that the history is in
order without any gaps (plus it was easier that way).

In the worst case, this will reset the entire history and slow rebuild
from there.

Note that we repair unmerged and merged histories separately.  This
can cause one series to have more patches than the other.  This is
mostly a non-issue as dak will cope with it (as long as both series
are consistent internally).
Signed-off-by: NNiels Thykier <niels@thykier.net>

This was suggested by the python-apt maintainer to work around #977000.

Signed-off-by: NNiels Thykier <niels@thykier.net>

Signed-off-by: NNiels Thykier <niels@thykier.net>

With the "daklib.daksubprocess" module gone, some modules no longer
saw the "daklib" modules, but tried to access other modules under it
which then failed.

Since Python 3.2 the standard subprocess modules restores the signal
handler to `SIG_DFL` by default.

Signed-off-by: NNiels Thykier <niels@thykier.net>

Signed-off-by: NNiels Thykier <niels@thykier.net>

Signed-off-by: NNiels Thykier <niels@thykier.net>