waf/waflib/Runner.py

#!/usr/bin/env python
# encoding: utf-8
# Thomas Nagy, 2005-2018 (ita)

"""
Runner.py: Task scheduling and execution
"""

import heapq, traceback
try:
	from queue import Queue, PriorityQueue
except ImportError:
	from Queue import Queue
	try:
		from Queue import PriorityQueue
	except ImportError:
		class PriorityQueue(Queue):
			def _init(self, maxsize):
				self.maxsize = maxsize
				self.queue = []
			def _put(self, item):
				heapq.heappush(self.queue, item)
			def _get(self):
				return heapq.heappop(self.queue)

from waflib import Utils, Task, Errors, Logs

GAP = 5
"""
Wait for at least ``GAP * njobs`` before trying to enqueue more tasks to run
"""

class PriorityTasks(object):
	def __init__(self):
		self.lst = []
	def __len__(self):
		return len(self.lst)
	def __iter__(self):
		return iter(self.lst)
	def __str__(self):
		return 'PriorityTasks: [%s]' % '\n  '.join(str(x) for x in self.lst)
	def clear(self):
		self.lst = []
	def append(self, task):
		heapq.heappush(self.lst, task)
	def pop(self):
		return heapq.heappop(self.lst)
	def extend(self, lst):
		if self.lst:
			for x in lst:
				self.append(x)
		else:
			if isinstance(lst, list):
				self.lst = lst
				heapq.heapify(lst)
			else:
				self.lst = lst.lst

class Consumer(Utils.threading.Thread):
	"""
	Daemon thread object that executes a task. It shares a semaphore with
	the coordinator :py:class:`waflib.Runner.Spawner`. There is one
	instance per task to consume.
	"""
	def __init__(self, spawner, task):
		Utils.threading.Thread.__init__(self)
		self.task = task
		"""Task to execute"""
		self.spawner = spawner
		"""Coordinator object"""
		self.setDaemon(1)
		self.start()
	def run(self):
		"""
		Processes a single task
		"""
		try:
			if not self.spawner.master.stop:
				self.spawner.master.process_task(self.task)
		finally:
			self.spawner.sem.release()
			self.spawner.master.out.put(self.task)
			self.task = None
			self.spawner = None

class Spawner(Utils.threading.Thread):
	"""
	Daemon thread that consumes tasks from :py:class:`waflib.Runner.Parallel` producer and
	spawns a consuming thread :py:class:`waflib.Runner.Consumer` for each
	:py:class:`waflib.Task.Task` instance.
	"""
	def __init__(self, master):
		Utils.threading.Thread.__init__(self)
		self.master = master
		""":py:class:`waflib.Runner.Parallel` producer instance"""
		self.sem = Utils.threading.Semaphore(master.numjobs)
		"""Bounded semaphore that prevents spawning more than *n* concurrent consumers"""
		self.setDaemon(1)
		self.start()
	def run(self):
		"""
		Spawns new consumers to execute tasks by delegating to :py:meth:`waflib.Runner.Spawner.loop`
		"""
		try:
			self.loop()
		except Exception:
			# Python 2 prints unnecessary messages when shutting down
			# we also want to stop the thread properly
			pass
	def loop(self):
		"""
		Consumes task objects from the producer; ends when the producer has no more
		task to provide.
		"""
		master = self.master
		while 1:
			task = master.ready.get()
			self.sem.acquire()
			if not master.stop:
				task.log_display(task.generator.bld)
			Consumer(self, task)

class Parallel(object):
	"""
	Schedule the tasks obtained from the build context for execution.
	"""
	def __init__(self, bld, j=2):
		"""
		The initialization requires a build context reference
		for computing the total number of jobs.
		"""

		self.numjobs = j
		"""
		Amount of parallel consumers to use
		"""

		self.bld = bld
		"""
		Instance of :py:class:`waflib.Build.BuildContext`
		"""

		self.outstanding = PriorityTasks()
		"""Heap of :py:class:`waflib.Task.Task` that may be ready to be executed"""

		self.postponed = PriorityTasks()
		"""Heap of :py:class:`waflib.Task.Task` which are not ready to run for non-DAG reasons"""

		self.incomplete = set()
		"""List of :py:class:`waflib.Task.Task` waiting for dependent tasks to complete (DAG)"""

		self.ready = PriorityQueue(0)
		"""List of :py:class:`waflib.Task.Task` ready to be executed by consumers"""

		self.out = Queue(0)
		"""List of :py:class:`waflib.Task.Task` returned by the task consumers"""

		self.count = 0
		"""Amount of tasks that may be processed by :py:class:`waflib.Runner.TaskConsumer`"""

		self.processed = 0
		"""Amount of tasks processed"""

		self.stop = False
		"""Error flag to stop the build"""

		self.error = []
		"""Tasks that could not be executed"""

		self.biter = None
		"""Task iterator which must give groups of parallelizable tasks when calling ``next()``"""

		self.dirty = False
		"""
		Flag that indicates that the build cache must be saved when a task was executed
		(calls :py:meth:`waflib.Build.BuildContext.store`)"""

		self.revdeps = Utils.defaultdict(set)
		"""
		The reverse dependency graph of dependencies obtained from Task.run_after
		"""

		self.spawner = None
		"""
		Coordinating daemon thread that spawns thread consumers
		"""
		if self.numjobs > 1:
			self.spawner = Spawner(self)

	def get_next_task(self):
		"""
		Obtains the next Task instance to run

		:rtype: :py:class:`waflib.Task.Task`
		"""
		if not self.outstanding:
			return None
		return self.outstanding.pop()

	def postpone(self, tsk):
		"""
		Adds the task to the list :py:attr:`waflib.Runner.Parallel.postponed`.
		The order is scrambled so as to consume as many tasks in parallel as possible.

		:param tsk: task instance
		:type tsk: :py:class:`waflib.Task.Task`
		"""
		self.postponed.append(tsk)

	def refill_task_list(self):
		"""
		Pulls a next group of tasks to execute in :py:attr:`waflib.Runner.Parallel.outstanding`.
		Ensures that all tasks in the current build group are complete before processing the next one.
		"""
		while self.count > self.numjobs * GAP:
			self.get_out()

		while not self.outstanding:
			if self.count:
				self.get_out()
				if self.outstanding:
					break
			elif self.postponed:
				try:
					cond = self.deadlock == self.processed
				except AttributeError:
					pass
				else:
					if cond:
						# The most common reason is conflicting build order declaration
						# for example: "X run_after Y" and "Y run_after X"
						# Another can be changing "run_after" dependencies while the build is running
						# for example: updating "tsk.run_after" in the "runnable_status" method
						lst = []
						for tsk in self.postponed:
							deps = [id(x) for x in tsk.run_after if not x.hasrun]
							lst.append('%s\t-> %r' % (repr(tsk), deps))
							if not deps:
								lst.append('\n  task %r dependencies are done, check its *runnable_status*?' % id(tsk))
						raise Errors.WafError('Deadlock detected: check the task build order%s' % ''.join(lst))
				self.deadlock = self.processed

			if self.postponed:
				self.outstanding.extend(self.postponed)
				self.postponed.clear()
			elif not self.count:
				if self.incomplete:
					for x in self.incomplete:
						for k in x.run_after:
							if not k.hasrun:
								break
						else:
							# dependency added after the build started without updating revdeps
							self.incomplete.remove(x)
							self.outstanding.append(x)
							break
					else:
						if self.stop or self.error:
							break
						raise Errors.WafError('Broken revdeps detected on %r' % self.incomplete)
				else:
					tasks = next(self.biter)
					ready, waiting = self.prio_and_split(tasks)
					self.outstanding.extend(ready)
					self.incomplete.update(waiting)
					self.total = self.bld.total()
					break

	def add_more_tasks(self, tsk):
		"""
		If a task provides :py:attr:`waflib.Task.Task.more_tasks`, then the tasks contained
		in that list are added to the current build and will be processed before the next build group.

		The priorities for dependent tasks are not re-calculated globally

		:param tsk: task instance
		:type tsk: :py:attr:`waflib.Task.Task`
		"""
		if getattr(tsk, 'more_tasks', None):
			more = set(tsk.more_tasks)
			groups_done = set()
			def iteri(a, b):
				for x in a:
					yield x
				for x in b:
					yield x

			# Update the dependency tree
			# this assumes that task.run_after values were updated
			for x in iteri(self.outstanding, self.incomplete):
				for k in x.run_after:
					if isinstance(k, Task.TaskGroup):
						if k not in groups_done:
							groups_done.add(k)
							for j in k.prev & more:
								self.revdeps[j].add(k)
					elif k in more:
						self.revdeps[k].add(x)

			ready, waiting = self.prio_and_split(tsk.more_tasks)
			self.outstanding.extend(ready)
			self.incomplete.update(waiting)
			self.total += len(tsk.more_tasks)

	def mark_finished(self, tsk):
		def try_unfreeze(x):
			# DAG ancestors are likely to be in the incomplete set
			# This assumes that the run_after contents have not changed
			# after the build starts, else a deadlock may occur
			if x in self.incomplete:
				# TODO remove dependencies to free some memory?
				# x.run_after.remove(tsk)
				for k in x.run_after:
					if not k.hasrun:
						break
				else:
					self.incomplete.remove(x)
					self.outstanding.append(x)

		if tsk in self.revdeps:
			for x in self.revdeps[tsk]:
				if isinstance(x, Task.TaskGroup):
					x.prev.remove(tsk)
					if not x.prev:
						for k in x.next:
							# TODO necessary optimization?
							k.run_after.remove(x)
							try_unfreeze(k)
						# TODO necessary optimization?
						x.next = []
				else:
					try_unfreeze(x)
			del self.revdeps[tsk]

		if hasattr(tsk, 'semaphore'):
			sem = tsk.semaphore
			try:
				sem.release(tsk)
			except KeyError:
				# TODO
				pass
			else:
				while sem.waiting and not sem.is_locked():
					# take a frozen task, make it ready to run
					x = sem.waiting.pop()
					self._add_task(x)

	def get_out(self):
		"""
		Waits for a Task that task consumers add to :py:attr:`waflib.Runner.Parallel.out` after execution.
		Adds more Tasks if necessary through :py:attr:`waflib.Runner.Parallel.add_more_tasks`.

		:rtype: :py:attr:`waflib.Task.Task`
		"""
		tsk = self.out.get()
		if not self.stop:
			self.add_more_tasks(tsk)
		self.mark_finished(tsk)

		self.count -= 1
		self.dirty = True
		return tsk

	def add_task(self, tsk):
		"""
		Enqueue a Task to :py:attr:`waflib.Runner.Parallel.ready` so that consumers can run them.

		:param tsk: task instance
		:type tsk: :py:attr:`waflib.Task.Task`
		"""
		# TODO change in waf 2.1
		self.ready.put(tsk)

	def _add_task(self, tsk):
		if hasattr(tsk, 'semaphore'):
			sem = tsk.semaphore
			try:
				sem.acquire(tsk)
			except IndexError:
				sem.waiting.add(tsk)
				return

		self.count += 1
		self.processed += 1
		if self.numjobs == 1:
			tsk.log_display(tsk.generator.bld)
			try:
				self.process_task(tsk)
			finally:
				self.out.put(tsk)
		else:
			self.add_task(tsk)

	def process_task(self, tsk):
		"""
		Processes a task and attempts to stop the build in case of errors
		"""
		tsk.process()
		if tsk.hasrun != Task.SUCCESS:
			self.error_handler(tsk)

	def skip(self, tsk):
		"""
		Mark a task as skipped/up-to-date
		"""
		tsk.hasrun = Task.SKIPPED
		self.mark_finished(tsk)

	def cancel(self, tsk):
		"""
		Mark a task as failed because of unsatisfiable dependencies
		"""
		tsk.hasrun = Task.CANCELED
		self.mark_finished(tsk)

	def error_handler(self, tsk):
		"""
		Called when a task cannot be executed. The flag :py:attr:`waflib.Runner.Parallel.stop` is set,
		unless the build is executed with::

			$ waf build -k

		:param tsk: task instance
		:type tsk: :py:attr:`waflib.Task.Task`
		"""
		if not self.bld.keep:
			self.stop = True
		self.error.append(tsk)

	def task_status(self, tsk):
		"""
		Obtains the task status to decide whether to run it immediately or not.

		:return: the exit status, for example :py:attr:`waflib.Task.ASK_LATER`
		:rtype: integer
		"""
		try:
			return tsk.runnable_status()
		except Exception:
			self.processed += 1
			tsk.err_msg = traceback.format_exc()
			if not self.stop and self.bld.keep:
				self.skip(tsk)
				if self.bld.keep == 1:
					# if -k stop on the first exception, if -kk try to go as far as possible
					if Logs.verbose > 1 or not self.error:
						self.error.append(tsk)
					self.stop = True
				else:
					if Logs.verbose > 1:
						self.error.append(tsk)
				return Task.EXCEPTION

			tsk.hasrun = Task.EXCEPTION
			self.error_handler(tsk)

			return Task.EXCEPTION

	def start(self):
		"""
		Obtains Task instances from the BuildContext instance and adds the ones that need to be executed to
		:py:class:`waflib.Runner.Parallel.ready` so that the :py:class:`waflib.Runner.Spawner` consumer thread
		has them executed. Obtains the executed Tasks back from :py:class:`waflib.Runner.Parallel.out`
		and marks the build as failed by setting the ``stop`` flag.
		If only one job is used, then executes the tasks one by one, without consumers.
		"""
		self.total = self.bld.total()

		while not self.stop:

			self.refill_task_list()

			# consider the next task
			tsk = self.get_next_task()
			if not tsk:
				if self.count:
					# tasks may add new ones after they are run
					continue
				else:
					# no tasks to run, no tasks running, time to exit
					break

			if tsk.hasrun:
				# if the task is marked as "run", just skip it
				self.processed += 1
				continue

			if self.stop: # stop immediately after a failure is detected
				break

			st = self.task_status(tsk)
			if st == Task.RUN_ME:
				self._add_task(tsk)
			elif st == Task.ASK_LATER:
				self.postpone(tsk)
			elif st == Task.SKIP_ME:
				self.processed += 1
				self.skip(tsk)
				self.add_more_tasks(tsk)
			elif st == Task.CANCEL_ME:
				# A dependency problem has occurred, and the
				# build is most likely run with `waf -k`
				if Logs.verbose > 1:
					self.error.append(tsk)
				self.processed += 1
				self.cancel(tsk)

		# self.count represents the tasks that have been made available to the consumer threads
		# collect all the tasks after an error else the message may be incomplete
		while self.error and self.count:
			self.get_out()

		self.ready.put(None)
		if not self.stop:
			assert not self.count
			assert not self.postponed
			assert not self.incomplete

	def prio_and_split(self, tasks):
		"""
		Label input tasks with priority values, and return a pair containing
		the tasks that are ready to run and the tasks that are necessarily
		waiting for other tasks to complete.

		The priority system is really meant as an optional layer for optimization:
		dependency cycles are found quickly, and builds should be more efficient.
		A high priority number means that a task is processed first.

		This method can be overridden to disable the priority system::

			def prio_and_split(self, tasks):
				return tasks, []

		:return: A pair of task lists
		:rtype: tuple
		"""
		# to disable:
		#return tasks, []
		for x in tasks:
			x.visited = 0

		reverse = self.revdeps

		groups_done = set()
		for x in tasks:
			for k in x.run_after:
				if isinstance(k, Task.TaskGroup):
					if k not in groups_done:
						groups_done.add(k)
						for j in k.prev:
							reverse[j].add(k)
				else:
					reverse[k].add(x)

		# the priority number is not the tree depth
		def visit(n):
			if isinstance(n, Task.TaskGroup):
				return sum(visit(k) for k in n.next)

			if n.visited == 0:
				n.visited = 1

				if n in reverse:
					rev = reverse[n]
					n.prio_order = n.tree_weight + len(rev) + sum(visit(k) for k in rev)
				else:
					n.prio_order = n.tree_weight

				n.visited = 2
			elif n.visited == 1:
				raise Errors.WafError('Dependency cycle found!')
			return n.prio_order

		for x in tasks:
			if x.visited != 0:
				# must visit all to detect cycles
				continue
			try:
				visit(x)
			except Errors.WafError:
				self.debug_cycles(tasks, reverse)

		ready = []
		waiting = []
		for x in tasks:
			for k in x.run_after:
				if not k.hasrun:
					waiting.append(x)
					break
			else:
				ready.append(x)
		return (ready, waiting)

	def debug_cycles(self, tasks, reverse):
		tmp = {}
		for x in tasks:
			tmp[x] = 0

		def visit(n, acc):
			if isinstance(n, Task.TaskGroup):
				for k in n.next:
					visit(k, acc)
				return
			if tmp[n] == 0:
				tmp[n] = 1
				for k in reverse.get(n, []):
					visit(k, [n] + acc)
				tmp[n] = 2
			elif tmp[n] == 1:
				lst = []
				for tsk in acc:
					lst.append(repr(tsk))
					if tsk is n:
						# exclude prior nodes, we want the minimum cycle
						break
				raise Errors.WafError('Task dependency cycle in "run_after" constraints: %s' % ''.join(lst))
		for x in tasks:
			visit(x, [])