Pkgcore

--------------------------

- filter-env grew a --print-vars option. If you've been seeing
"declare: write error: Broken pipe" from build operations, this should
now be fixed via using this new option.

- the resolver wasn't properly accounting for planned modifications to
the installed pkgs database. If you've had upgrade issues from
blockers, this is the root cause (pam/pambase in particular).

- eclass scanning is now JIT'd, and the resultant eclass dictionary
is now marked immutable for safety reasons.

- for portage configuration when PORTDIR_OVERLAY is in use and portdir
has a pregenerated cache, check the pregenerated cache first when
looking for metadata. This degrades the usage case where overlays
override quite a few core eclasses in favor of the more common case
where the pregenerated cache is the majority of the time, accurate.
End result is upwards of a 2x reduction in open invocations.


--------------------------

--------------------------

- touch vdb root on vdb modification as a way to notify alternative PMs
that their cache needs updating. Gentoo bug 290428. Just leaves paludis
to join in on the fun...

- portage 2.2 modified make.globals to add a default, non glep23 compliant
ACCEPT_LICENSE. pkgcore's implementation has been modified to be non
compliant to glep23, matching portage semantics.

If pquery <atom> has suddenly started returning nothing, this was the cause.

- fix a traceback that could occur when doing pmerge -pv for when binpkg
repos were involved.


--------------------------

--------------------------

- correct a python-2.6 incompatibility that rears its head when doing
repository operations (installing, uninstalling, etc).


------------------------

------------------------

- add protection against multiple python versions, w/ the default python
invocation being a different major version from what pkgcore was installed
under. Primarily a fix to dohtml.

- ticket 230; tweaks for better >=python2.5 compatibility.

- pkgcore will now try to sync overlays if the overlay is a vcs. This can
be disabled by adding FEATURES="-autodetect-sync" to your make.conf

- pkgcore.sync.base.AutodetectSyncer was added as a way to pull configuration
from existing on disk vcs repos, and generate a syncer from them.

- handle cache corruption a bit better- namely, log the warning, and keep
going. Degradation of performance can result, but it's preferable to just
bailing.

- gentoo bug 280766; basically some ebuilds are sensitive to a trailing '/'
on WORKDIR (portage strips it) leading to failures in path sedding.

- comply with PMS corner cases for package names; gentoo bug 263787

- serialization support for cpv derivatives. Not great, but packages.g.o
relies on it, thus its inclusion.

- not surprising on the timing (or spotting it via ciaran spreading it
via blog comments), gentoo bug 226505 revisited- change in phase ordering
afflicting all eapis. pkgcore had it right the first time, inverted the
ordering in 0.4.7.9.


-----------------------------

-----------------------------

- pmerge is a bit more informative when there is nothing to merge,
and doesn't ask if in --ask if the users wishes to proceed.
Thanks to DJ Anderson for pointing out this oversight.

- ensure unicode for pquery --attr longdescription w/in pquery; via this
it leaves the unicode question to the formatter, instead of down converting
earlier.

- fix a mismatch between src ebuilds and binpkgs for _eclasses_ when
doing pquery --attr inherited. Bit of a hack, but it'll suffice.

- pquery --attr all and --attr allmetadata was added. 'all' gets you
all known attrs (environment, contents, etc); bit heavy but useful if
you need to see it all. 'allmetadata' gets you the key/val pairs for
this host- fetchables, depends, slotting, eapi, repo, cbuild/chost, etc,
but no environment/contents.

- fix cycle detection for dev-util/git; specifically there is a cycle on
virtual/perl-Module-Built which can be ignored since that chain of deps
are pulled in via post_rdepends.

- make gid/mode configurable for filelist pkgsets; this fixes 4 failures
for when the tests are ran and the user isn't a member of portage.

- fix a cornercase in fs.livefs.intersect where intersecting a file/dir
would trigger a traceback.

- fix a corner case where the world file isn't updated if the world file
is empty.

- fix a deprecation warning under 2.6 caused by an impedence between
native_PackageRestriction and the cpy version for __init__ invocation.

- fix gentoo bug 216492, a change in libsandbox behaviour- specifically
libsandbox for >=1.3 is now appending libsandbox.so while failing to
spot it already existing in LD_PRELOAD; pkgcore tests were written a bit
strict thus were spotting this. Loosen the test, and fix the case where
a different preload is used in conjunction w/ sandbox.


-----------------------------

-----------------------------

- fix docutils-0.5 incompatibility in build_api_docs.py

- python issue 4230 makes __getattr__ support descriptor protocol.
This unfortunately causes part of config handling to go boom- fixed.

Unfortunately this also means that we need to support both descriptor
and *non* descriptor interpretters at *runtime*- if python is upgraded
underfoot, things get unhappy to keep atom.__getattr__ from blowing up.
Fixed either way.

- copy HOMEPAGE into vdb/binpkg by default.


-----------------------------