Pkgcore

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

- merging/replacing performance may be a bit slower in this release- the level
of stats calls went up in comparison to previous releases, with several
duplicates. This will be corrected in the next release- releasing in the
interim for bugfixes this version contains.

- add CBUILD=${CBUILD:-${CHOST}}; couple of odd ebuilds rely on it despite
being outside of PMS.

- protective trigger was added blocking unmerging of a basic set of
directories/syms; mainly /*, and /usr/*.

- when a merge passes through a symlink for path resolution, that sym is
no longer pulled in as an entry of that pkg. Originally this was done for
protective reasons, but it serves long term as a way to inadvertantly hold
onto undesired junk from the users fs, and opens the potential to unmerge
system/global symlinks when that pkg/slot's refcount hits zero.

- detection, and predicting merge locations for syms was doing an unecessary
level of stat calls; this has been reduced to bare minimum.

- ticket 159; force an realpath of CONTENTS coming from the vdb due to other
managers not always writing realpath'd entries, thus resulting in occasional
misidentification of what to remove.

- pkgcore.util.parserestrict no longer throws MalformedAtom, always
ParseError. Removes ugly commandline tracebacks for bad atoms supplied
to pmerge.

- ticket 158; honor RSYNC_PROXY for rsync syncer.
Thanks to user Ford_Prefect.

- pmerge -N now implies --oneshot.

- correct a flaw in tbz2 merging where it repeatedly try to seek in the bz2
stream to generate chksums, instead of using the on disk files for
chksumming.

- pmaint regen w/ > 1 thread no longer throws an ugly set of tracebacks upon
completion.

- binpkg repositories now tell you the offending mode, and what is needed
to correct it. No longer cares if the specified binpkg base location is
a symlink also.

- pmaint --help usage descriptions are far more useful now.


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

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

- correct a corner case where a users bash_profile is noisy, specifically
disable using $HOME/.bashrc from all spawn_bash calls.

- USE=-* in make.conf support is restored. ticket 155.

- minor tweak to package.keywords, package.use, and package.license support-
-* is properly supported now. Following portage, if you're trying to
match keywords for a pkg that are '-* x86', you must match on x86.

- pquery --attr use output for EAPI=1 default IUSE is significantly less
ugly.

- ticket 150. EAPI1 IUSE defaults fixups. stacking order is that default
IUSE is basically first in the chain, so any configuration (global, per
pkg, etc), will override if possible. Effectively, this means a default
IUSE of "-foon" is pointless, since there is no earlier USE stack to
override.

- pkgcore.ebuild.collapsed_restrict_to_data api was broken outside of a
major version bump- specifically pull_cp_data method was removed since
the lone consumer (pkgcore internals) doesn't need it, and the method
is semi dangerous to use since it only examines atoms.


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

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

- correct handling of ebuilds with explicit -r0 in filename, despite it being
implicit. Thanks to rbrown for violating gentoo-x86 policy out of the blue
w/ an ebuild that has -r0 explicit in the filename for smoking out a bug
in pkgcore handling of it. Ebuild since removed, but the KeyError issue
is corrected. (keep the bugs coming)

- minor performance optimization to binpkg merging when there is a large
of symlink rewrites required.

- ticket 153; restore <0.4 behaviour for temporal blocker validation, rather
then invalidly relying on the initial vdb state for blocker checks. Fixes
resolution/merging of sys-libs/pam-0.99.10.0


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

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

- add tar contentsSet rewriting; tarballs sometimes leave out directories,
and don't always have the fully resolved path- /usr/lib/blah, when
/usr/lib -> /usr/lib64 *should* be /usr/lib64/blah, but tar doesn't force
this. Due to that, can lead to explosions in unpacking- this is now fixed.

- pquery --attr inherited was added; this feature may disappear down the
line, adding it meanwhile since it's useful for ebuild devs.

- adjust setup.py so that man page installation properly respects --root

- correct a corner case where a package name of 'dev-3D' was flagged as
invalid.


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

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

- resolver fixes: vdb loadup wasn't occuring for old style virtuals for
rdepend blockers, now forces it. It was possible for a node to be
considered usable before its rdepends blockers were leveled- now those
must be satisfied before being able to dep on the node.

- resolver events cleanup; pmerge now gives far better info as to why a
choice failed, what it attempted to get around it, etc.

- multiplex trees now discern their frozen state from their subtrees,
and will execute the repo_op for the leftmost subtree if unfrozen.

- pquery --attr eapi was added.

- ticket 94; package.provided is now supported fully both in profiles,
and in user profile (/etc/portage/profile).

- ticket 116; ignore empty tarfile exception if the exception explicitly
states empty header.

- utter corner case compatibility- =dev-util/diffball-1.0-r0 is now the
same as =dev-util/diffball-1.0 .

- convert FETCHCOMMAND/RESUMECOMMAND support to execute spawn_bash by
default instead of trying to cut out shell; this kills off the occasional
incompatibility introduced via portage supplying make.globals.

- FEATURES=sfperms is now a trigger instead of a dyn_preinst hook.
Faster, cleaner, etc.

- delayed unpacking of binpkgs has been disabled; occasionally can lead to
quadratic behaviour in contents accessing, and extreme corner case trigger
breakages. Will be re-enabled once API has been refactored to remove
these issues.

- FEATURES=multilib-strict was converted into a trigger. Tries to
use the python bindings for file first (merge file[python]), falling
back to invoking file. Strongly suggested you have the bindings- fair bit
faster. Finally, verification now runs for binpkgs also.

- bug 137; symlink on directory merging failures where pkgcore would wipe
files it had just installed invalidly.

- correct issue in offset rewriting (was resetting new_offset to '/')-
should only be api visible, no existing consumers known.

- ebuild env lzma unpack support was broken; fixed (ticket 140).

- Additional debug output for pmerge.

- Further extending PortageFormatter to sanely handle worldfile highlights
and show repos with both id and location

- Ticket 132: Portage Formatter supports real portage colors now,
thanks to agaffney for getting the ball rolling

- Masked IUSEs were not treated right in all cases, thanks to agaffney
for report and help testing

- diefunc tracebacks beautified


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

- expand repository api slightly adding has_match; this is intended
as a simple boolean check if a repo has it. It should *only* be
used for containment- if you need the results don't test then itermatch,
just itermatch.

- add cpy implementation of PackageRestriction.match

- for package.provided repositories, short circuit their itermatch/match
if there aren't any results possible.

- re-enable cpython implementation of DepSet parsing for eapi2- roughly
a 31% speedup for current gentoo-x86 repository dependency parsing.

- performance improvements to pquery --attr alldepends; specifically
depset.stringify_boolean is now 20% faster.

- performance improvements to pquery --attr alldepends -v


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