Update Portage to sync BlockerDB at init, rather than before every package.

pym/_emerge/Scheduler.py

Index: pym/_emerge/Scheduler.py
diff --git a/pym/_emerge/Scheduler.py b/pym/_emerge/Scheduler.py
index fa4825cf2058bbbe24ba4690a61a95a332315b32..e4b6926c3a110150d824b42fdd2f5a812b39196d 100644
--- a/pym/_emerge/Scheduler.py
+++ b/pym/_emerge/Scheduler.py
@@ -317,10 +317,6 @@ class Scheduler(PollScheduler):
 		Initialization structures used for dependency calculations
 		involving currently installed packages.
 		"""
-		# TODO: Replace the BlockerDB with a depgraph of installed packages
-		# that's updated incrementally with each upgrade/uninstall operation
-		# This will be useful for making quick and safe decisions with respect
-		# to aggressive parallelization discussed in bug #279623.
 		self._set_graph_config(graph_config)
 		self._blocker_db = {}
 		for root in self.trees:
@@ -329,6 +325,7 @@ class Scheduler(PollScheduler):
 					pkg_cache=self._pkg_cache)
 			else:
 				fake_vartree = graph_config.trees[root]['vartree']
+			fake_vartree.sync()

[diandersAtChromium, 2011/03/22 02:57:59]

How did you decide this was the place to put the sync()?

...this is not the only construction place for FakeVartree.  Do those other
places need sync() also?


To answer my own question:
* The only place that changed to not auto-call sync was findInstalledBlockers.
* The only caller of findInstalledBlockers is based on iterating over
self._blocker_db
* This is the only place that puts things in self._blocker_db.

...so I guess you're good?

[davidjames, 2011/03/23 20:19:11]

Exactly right.

 			self._blocker_db[root] = BlockerDB(fake_vartree)
 
 	def _destroy_graph(self):
@@ -643,27 +640,20 @@ class Scheduler(PollScheduler):
 
 	def _find_blockers(self, new_pkg):
 		"""
-		Returns a callable which should be called only when
-		the vdb lock has been acquired.
+		Returns a callable.

[diandersAtChromium, 2011/03/22 02:57:59]

...that will find the blockers for the given package.

[davidjames, 2011/03/23 20:19:11]

Done.

 		"""
 		def get_blockers():
-			return self._find_blockers_with_lock(new_pkg, acquire_lock=0)
+			return self._find_blockers_impl(new_pkg)
 		return get_blockers
 
-	def _find_blockers_with_lock(self, new_pkg, acquire_lock=0):
+	def _find_blockers_impl(self, new_pkg):
 		if self._opts_ignore_blockers.intersection(self.myopts):
 			return None
 

[diandersAtChromium, 2011/03/22 02:57:59]

Are you sure that the gc.collect() is no longer needed?  Seems like a different
CL should remove this, maybe (here and below).

[davidjames, 2011/03/23 20:19:11]

This gc.collect() call was only needed to workaround bugs in Python 2.5. Portage
has since dropped support for Python 2.5 so this is no longer applicable.

Not sure whether this needs to be split up into a separate CL.

-		# Call gc.collect() here to avoid heap overflow that
-		# triggers 'Cannot allocate memory' errors (reported
-		# with python-2.5).
-		gc.collect()
-
 		blocker_db = self._blocker_db[new_pkg.root]
 
 		blocker_dblinks = []
-		for blocking_pkg in blocker_db.findInstalledBlockers(
-			new_pkg, acquire_lock=acquire_lock):
+		for blocking_pkg in blocker_db.findInstalledBlockers(new_pkg):
 			if new_pkg.slot_atom == blocking_pkg.slot_atom:
 				continue
 			if new_pkg.cpv == blocking_pkg.cpv:
@@ -673,8 +663,6 @@ class Scheduler(PollScheduler):
 				self.pkgsettings[blocking_pkg.root], treetype="vartree",
 				vartree=self.trees[blocking_pkg.root]["vartree"]))
 
-		gc.collect()
-
 		return blocker_dblinks
 
 	def _dblink_pkg(self, pkg_dblink):
@@ -1527,6 +1515,8 @@ class Scheduler(PollScheduler):
 		self._completed_tasks.add(pkg)
 		self._unsatisfied_system_deps.discard(pkg)
 		self._choose_pkg_return_early = False
+		blocker_db = self._blocker_db[pkg.root]
+		blocker_db.discardBlocker(pkg)
 
 	def _merge(self):