Scenario:
dependency = dev-lang/python
SLOT intersection = enabled
available slots = 2.6, 2.7, 3.2
installed slot = 2.7
The SLOT intersection feature worked correctly but in the end
dev-lang/python was added to the unsatisfied set because there
was a dev-lang/python-2.7 update as well. This results in other
dep calculation code to match dev-lang/python against 3.2 instead.
The SLOT intersection feature has to reduce the dependency string
scope by adding a SLOT suffix (in this case: ":2.7"). This way the
correct dependency is eventually pulled in.
Always flush stderr on a newline and also write the number of
active threads as well.
Moreover, always install the dump function at SIGQUIT when --debug
is passed.
This makes possible to have external module packages inside the
sys-kernel/ category, by just adding extra checks. In particular,
we now check against ETYPE ebuild environment variable.
This should guarantee that all the kernel sources, binaries and
headers won't be tagged by Entropy.
Everytime Repository.sync() is triggered, the last sync time
information is updated. This makes possible to figure out when
the last repository update has been run and warn user about it
if they're too old.
This makes possible to support the "readers/writers" model by
letting code to acquire Entropy Resources in shared mode (for
readers) and the same in exclusive mode (default, for writers).
In order to simplify the code and allow further improvements
(eg. the addition of readers/writers locking), migrate all the
Entropy Resources locking functions to FlockFile().
Use a semaphore to signal the worker thread if there's something to do,
otherwise the whole thread should just sleep and avoid to poll every
N seconds.
ebuild environment setup function for pkg_{prerm,postrm} can raise
OSError during os.makedirs(). Better catch it and handle the failure
nicely so that the user can take proper measures to fix it, instead
of completely bailing out with garbageish text.
It can happen that HTTP servers (or those accessed via Proxy) do not
actually provide Content-Length information in the response.
This caused UrlFetcher to keep __remotesize = 0, which means that
the same thought that the file has been fully downloaded already
under some circumstances.
As of today, both Equo and Sulfur implemented their own search
functions because the API didn't provide any direct support for
searching through all the repositories. atom_search() now offers
a basic way to do this, without reinventing the weel every time
in consumer code.
If the installed packages repository has been regenerated manually,
some entries may contain invalid digest values. Given this, the same
value is forced to be equal to "0" so that we can skip the check during
dep calculation if an entry like that is found.
Using pragma inside executescript() won't work and foreign_keys were
kept enabled, resulting in extrainfo table being wiped.
At the same time, dropping all the indexes before starting dramatically
improves the execution speed.
Graph caching must hold copies of cache objects instead of giving
them away for free and risk of having them polluted or even completely
cleared out under its own chair.
Example:
$ equo remove linux-sabayon:3.1 [--deep]
The --deep version doesn't pull in all the reverse deps due to
Graph.solve() being called twice in
entropy.client.interfaces.dep's _generate_reverse_dependency_tree()
when deep = True.
Fixes bug #2989
Fabio Erculiani
Sławomir Nizio