PackageUninstall: return early if pkg is missing

PackageUninstall: remove extra cat/pf var init

Also, add to comments about ebuild-locks.

test_merge_order: test asap PDEPEND (bug #180045)

treewalk: fix PORTAGE_BUILDIR_LOCKED state

This fixes breakage from commit
f0f1bbe8fa9d3f698cbe529d2a11eec1ce437119.

PackageUninstall: log more output

PackageUninstall: make async with MergeProcess

This fixes another ebuild-locks issue like the one fixed in commit
a81460175a441897282b0540cefff8060f2b92dc, but this time we use a
subprocess to ensure that the ebuild-locks for pkg_prerm and
pkg_postrm do not interfere with pkg_setup ebuild-locks held by
the main process.

doebuild_environment: move EAPI code later

All EAPI dependent code comes last, so that essential
variables like PORTAGE_BUILDDIR are still initialized even in
cases when UnsupportedAPIException needs to be raised, which
can be useful when uninstalling a package that has corrupt
EAPI metadata.

MergeProcess: remove unnecessary dblink attribute

PackageUninstall: use _unmerge_display()

This is preparation for asynchronous support.

unmerge: split out _unmerge_display() function

This will be useful for adding asynchronous support to the
PackageUninstall class, since it will be able to call
_unmerge_display() for the display portion, and then do the
actual unmerge asynchronously.

repoman: pass type_name to Package constructor

test_merge_order: add more asap libc deps

_serialize_tasks: prefer unsatisfied asap child

Optimally, satisfied deps are always merged after the asap nodes that
depend on them.

test_depth: use merge_order_assertions

_add_pkg_dep_string: handle satisfied virt_dep

This fixes a problem with the asap LIBC_PACKAGE_ATOM test case for
bug #303567.

test_merge_order: test asap LIBC_PACKAGE_ATOM

_validate_blockers: reference bug 128809

test_merge_order: test asap PORTAGE_PACKAGE_ATOM

test_merge_order: test case for bug 250286

test_merge_order: test hard block resolution

test_merge_order: test many-to-one blocks

Package: add new _gen_hash_key method

test_depth: use ambiguous_merge_order

ambiguous_merge_order: set match = False more

Fix spelling of "ambiguous".

DepPriority: remove "rebuild" attribute

Since the addition of DepPriorityNormalRange and
DepPrioritySatisfiedRange in commit
bd369956b2a2fbc019a655a372628998499156c0, which solves most cases of
bug 199856, the Depriority.rebuild attribute doesn't appear to make any
difference. The edges that this attribute differentiates are already
naturally differentiated by the fact that the child node of a satisfied
buildtime dependency that's not being rebuilt will naturally be
identified as a leaf node earlier and removed from the graph, thereby
eliminating the edge before there's an opportunity to compare it with
a higher priority rebuild edge. The addition of the "optional"
attribute (in commit 15476805a156acd11fdaaa19212691e8ee09b309) also
plays a role here, since it converts some satisfied buildtime edges to
optional edges, thereby reducing their priority.

test_merge_order: fix bugs in previous commit

test_merge_order: test solve blocker no-overlap

treewalk: force AUTOCLEAN=yes if libs preserved

_prune_plib_registry: correct _match_contents use

merge_order_assertions: fix string format error

test_merge_order: test satisfied circular DEPEND

test_merge_order: test unsolvable buildtime cycle

test_merge_order: test circular RDEPEND + DEPEND

test_merge_order: make cycles indirect

Task.__eq__: allow _hash_key as input

This is used by depgraph._pkg() for lookups.

ambigous_merge_order: handle more failure modes

test_merge_order: test hard buildtime blockers

test_merge_order: test hard blocker resolution

test_merge_order: test blocker resolution

test_merge_order: remove uneeded ambigous flag

test_merge_order: fix ambigous_merge_order bug

test_merge_order: test circular RDEPEND + PDEPEND

ResolverPlayground: support ambigous_merge_order

Fix grammar in previous commit.

depclean: adjust failure message wrt --with-bdeps

We need to be more explicit about using --with-bdeps=y, since this
subtlety often leads to confusion.

Task: optimize == and != _hash_key comparison

Add FEATURES=distcc-pump support.

This will fix bug #288597.

Task: precalculate _hash_value

Task: remove unnecessary _get_hash_key() method

Package: precalculate _hash_key

This eliminates an expensive getattr call in _get_hash_key(), which
greatly improves depgraph performance on a weak ARM cpu.

Fix typos in comments.

_find_libs_to_preserve: always return set

preserve-libs: check for broken linkmap more

_pkg_visibility_check: if in graph then visible

_LockProcess.unlock: assert successful returncode

etc-update: clear term when appropriate

This solves an issue like bug #142508, but involving etc-update instead
of dispatch-conf. This is also relevant to bug #70668, which is the
same issue, though the reporter suggested to use colors to delimit the
output instead of using clear.

--autounmask-write: use CONFIGROOT more

Fix a typo in a comment.

digraph: implement __bool__

depgraph._resolve: use temp set for autounmask

display_autounmask: optimize check_if_latest

test_asynchronous_lock: simulate SIGINT from tty

AsynchronousTask: tweak poll() logic

AsynchronousLock: implement _poll() and test

test_asynchronous_lock: test returncode more

test_asynchronous_lock: test wait and cancel

test_asynchronous_lock: test waiting

test_autounmask: test missing keyword

Update timestamps in headers of modified files.

--autounmask-write: don't write hidden files

--autounmask-write: use CONFIGROOT and EROOT

--autounmask-write: handle non-existent file

--autounmask-write: only prompt when necessary

--autounmask-write: handle config file recursion

Document that --autounmask is enabled by default.

Enable --autounmask by default

--autounmask-write: respect --ask

BinpkgFetcher: handle AsynchronousLock failure

EbuildBuildDir: handle AsynchronousLock failure

_LockProcess: handle process failure more

_LockProcess: handle process failure if cancelled

preserve-libs: lock vardb for removeFromContents

Revert "counter_tick_core: don't lock if parallel-install"

This reverts commit 405ad9eed65393205ec28af8772f7ea45ce0371e.
The root problem is fixed by commit
8380ec9d7590a47b912e0d1e1db2ebb6049b903b.

treewalk: unlockdb/lockdb typo triggers EDEADLK

make.conf.txt: remove because unused/unmaintained

Fix some typos.

--autounmask: Treat missing keywords as masks

--autounmask: Allow package.mask changes

Implement --autounmask-write

Enabling this option together with --autounmask writes proposed
changes to config files, honoring CONFIG_PROTECT.

counter_tick_core: don't lock if parallel-install

This is the same as commit 461564ae94ff936918eeaa18493bc1da3846796f
but this time with comments that make sense.

EbuildMerge: inherit from CompositeTask

EbuildPhase: fix locking for setup phase

EbuildPhase: AsynchronousLock for ebuild-locks

Use vardbapi.lock() where applicable.

movefile: handle EEXIST when os.symlink fails

Revert "counter_tick_core: don't lock if parallel-install"

This reverts commit 461564ae94ff936918eeaa18493bc1da3846796f.
Disabling the lock here doesn't really make sense, and I'm still
triggering EDEADLK is some other places.

preserve-libs: use vardbapi _fs_lock/unlock

The preserve-libs code is too dependent on the filesystem state to
allow merging/unmerge/env_update to execute concurrently.

vardbapi: add reentrant _fs_lock/unlock methods

_LibGraphNode: re-use the _key attribute

This allows us to avoid repeating any previous stat calls, which helps
to avoid potential race conditions due to inconsistent stat results
when the file system is being modified concurrently.