Nit Actor Model, with some examples by BlackMinou · Pull Request #2361 · nitlang/nit

BlackMinou · 2017-02-06T16:58:57Z

A better version than the previous PR !

The model injection and support module generation are now two phases in the frontend.

Needs #2357 to be able to compile with nitc, for now you have to use it as nitc module.nit -m actors_module.nit for compiling.

privat

good job

privat · 2017-02-08T16:30:24Z

+# Abstraction of an actor
+# It has a mailbox, can receive and process messages asynchronously
+abstract class Actor
+		super Thread


privat · 2017-02-08T16:41:54Z

 	fun wait(mutex: NativePthreadMutex) `{ pthread_cond_wait(self, mutex); `}
 end

+


why a blank line?

privat · 2017-02-08T20:06:52Z

+blocking, since it's direcly canceling the native pthread associated to the actor.
+
+For now, there isn't any mecanism to recreate and actor after it was terminated.
+Sending messages after terminating it results in unknown behaviour.


s/unknown/unspecified/

BlackMinou · 2017-02-15T20:08:17Z

test osx please

xymus · 2017-02-16T19:19:04Z

test osx please

xymus

I expect the use of pthreads::extra to break compatibility with macOS and maybe Android. The module doc is Offers some POSIX threads services that are not available on all platforms... While Android is not a priority you may have to add your tests to the darwin.skip blacklist file.

xymus · 2017-02-16T18:24:12Z

+
+## Waiting for all actors to finish processing
+
+Let's imagine you create a whole bunch of actors and make them do things asynchronously from the main thread. Your don't want your program to exit right after giving work to your actors.


s/Your/You/

The two sentences on the same line look like an oversight too.

xymus · 2017-02-16T18:25:30Z

+## Examples
+
+You can find example of differents small programs implemented with Nit actors in the `/examples`
+directory. For a really simple example, you can check `/examples/simple`.


Are these relative paths? Beginning with a / usually means the root of the FS or maybe of the depository.

xymus · 2017-02-16T18:27:18Z

+		end
+	end
+
+	# Ends `self`, not responsible if work was ongoing


"not responsible" does not say much, it looks like it cancels ongoing work. Or do you mean that it's dangerous?

xymus · 2017-02-16T18:36:23Z

+end
+
+# A Blocking queue implemented from a `ConcurrentList`
+# Corresponds to the mailbox of an actor


You should list which (3?) services are blocking. Other services like [], first and insert are not blocking nor do they unblock shift.

If you want to make this a public service, you should probably also implement pop, it's the opposite to push.

I don't really want people to use pop on a BlockingQueue, and this one is specific to my actors.

Ultimately, a future PR would be needed to abstract the concept of a BlockingQueue in the pthreads module I think. But there is enough things in this one to report this to a future one.

xymus · 2017-02-16T18:38:42Z

+		mutex.unlock
+	end
+
+	# If the real collection is empty, we wait


Or you could precise here what stops the wait, like:

If empty, blocks until an item is inserted with `push` or `unshift`

xymus · 2017-02-16T18:59:16Z

+	super Phase
+
+	# List of actor classes
+	var actors = new Array[String]


I'm nitpicking now but the doc pretty much repeat the line below, may I suggest a tweak adding an important detail "Source code of the actor classes to generate"?

xymus · 2017-02-16T19:00:11Z

+			# Return signature
+			var ret = ""
+			# Return statement
+			var ret2 = ""


The comments are so clear, yet the variable names are not...

Haha, I must have inherited some habits from @privat

xymus · 2017-02-16T19:00:41Z

+""")
+
+
+# The actual proxy call


Indentation

xymus · 2017-02-16T19:05:05Z

+
+		var module_path = "{mmodule_path}/{module_name}.nit"
+
+		var nit_module = new NitModule(module_name + " is no_warning(\"missing-doc\")")


You should set the modules annotations with nit_module.annotations.add "no_warning...".

xymus · 2017-02-16T19:05:50Z

 	# @toimplement
 	fun process_nmodule(nmodule: AModule) do end

+


xymus · 2017-02-16T19:37:51Z

test osx please

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

xymus

+1 and it looks OK on macOS after all.

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

A better version than the previous PR ! The model injection and support module generation are now two phases in the frontend. Needs #2357 to be able to compile with `nitc`, for now you have to use it as `nitc module.nit -m actors_module.nit` for compiling. Pull-Request: #2361 Reviewed-by: Alexis Laferrière <alexis.laf@xymus.net> Reviewed-by: Jean Privat <jean@pryen.org>

tremblay-guy · 2017-03-17T11:55:39Z

+		for i in [0..p.length[ do p[i] = i
+
+		var i = count.length - 1
+		while i > 0 do


Est-ce possible de faire une boucle à l'envers en nit -- au lieu d'un while:

for i in [count.length-1..0].step(-1) do

oui, avec un reverse_iterator enfait

tremblay-guy · 2017-03-17T12:03:26Z

+		var first = p[0]
+		if p[first] != 0 then
+			p.copy_to(0, pp.length, pp, 0)
+			loop


Est-ce qu'un simple while ne serait pas suffisant ici (avec la condition plus haut p[first] != 0 (donc pas besoin du loop et du break):

while p[first] != 0 do
...
end

les deux conditions sont différentes, l'une est sur p, et l'autre sur pp. Mais je peux quand même changer la boucle :)

tremblay-guy · 2017-03-17T12:07:28Z

+			if p[0] != 0 then
+				var flips = count_flips
+				maxflips = maxflips.max(flips)
+				if i % 2 == 0 then


chk_sum += i % 2 == 0 ? +flips : -flips

on a pas exactement d'opérateur ternaire comme cela en Nit, mais il y a une solution qui ressemble, je vais la faire !

tremblay-guy · 2017-03-17T12:08:24Z

+		var chk_sum = 0
+
+		var i = idx_min
+		loop


Je n'aime pas les loop/break quand un while ou for suffit:

for i in [idx_min..idx_max] do
if i != idx_min then next_permutation

tremblay-guy · 2017-03-17T12:14:18Z

+		for i in [0..n[ do count.add(0)
+
+		var task = 0
+		loop


while (task = task_id.get_and_increment) < n_tasks do
run_task(task)
end

on peut pas faire d'assignation dans un while en Nit malheureusement !

tremblay-guy · 2017-03-17T12:19:28Z

+			p.copy_to(0, i+1, pp, 0)
+
+			for j in [0..i] do
+				if j + d <= i then


p[j] = (j+d <= i) ? pp[j+d] : pp[j+d-i-1]

tremblay-guy · 2017-03-17T12:21:09Z

+chk_sums = new Array[Int].with_capacity(n_tasks)
+for i in [0..n_tasks[ do chk_sums.add(0)
+
+var actors = new Array[FannkuchRedux].with_capacity(8)


C'est quoi cette constante magique 8? => Constante symbolique!!

un oubli de ma part, c'est nb_actors

tremblay-guy · 2017-03-17T12:22:11Z

+			var zi2 = cib[y]
+
+			var b = 0
+			for j in [0..49[ do


Constante magique 49 => constante symbolique!

c'est le nombre d'itérations qu'on effectue pour déterminer si le nombre appartient à l'ensemble de mandelbrot, je vais le mettre en constante symbolique

tremblay-guy · 2017-03-17T12:24:29Z

+				zr2 = nzr2
+				zi2 = nzi2
+
+				if zr1 * zr1 + zi1 * zi1 > 4.0 then b = b | 2


Ceci me semble plus simple et clair:

if zr1 * zr1 + zi1 * zi1 > 4.0 && zr2 * zr2 + zi2 * zi2 > 4.0 then
break
end

Serait encore plus simple avec une méthode auxiliaire...

Par contre, je constate que b est utilisé pour r... mais d'une façon que je ne comprends pas trop :(

J'avoue que je ne comprends pas trop non plus cette partie de l'algo, je préfère ne pas y toucher ...

tremblay-guy · 2017-03-17T12:25:27Z

+
+	fun work do
+		var line = 0
+		loop


while (line = atomic.get_and_increment) < n do
put_line( line, data[line] )
end

idem que pour fannkuch, on ne peut pas faire d'assignation dans un while en Nit, je vais quand même remettre le if sur une ligne par contre

tremblay-guy · 2017-03-17T12:26:03Z

+	var nb_threads = 8
+end
+
+if args.is_empty then


n = args.is_empty ? 200 : args[0].to_i

tremblay-guy · 2017-03-17T12:26:56Z

+	print chk.to_s + "\nPfannfuchen(" + n.to_s + ") = " + res.to_s
+
+end
+


n = args.is_empty ? 7 : args[0].to_i

tremblay-guy · 2017-03-17T12:27:26Z

+else
+	n = args[0].to_i
+end
+


Pourquoi le +7?

bonne question, ça a l'air de fonctionner bien sans !

tremblay-guy · 2017-03-17T12:31:04Z

+	var samecount = 0
+
+	fun run do
+		loop


var p
while (p = place.meet(id, color)) != null do

toujours pas d'assignation dans les while :p

tremblay-guy · 2017-03-17T12:33:39Z

+	var colors: Array[String] = ["blue", "red", "yellow"]
+end
+
+fun complement(c, other: Int): Int do


Ouille. Pas moyen de faire ça avec une table?

j'ait fait un Array[Array[Int]] à la place, mais c'est pas plus perf !

tremblay-guy · 2017-03-17T12:35:10Z

+fun get_number(n: Int): String do
+	var str = ""
+	var nstr = n.to_s
+	for c in nstr do


Est-ce que "+" en Nit crée une nouvelle chaine? (Comme en Java et Ruby?)

Est-ce qu'il existe un append comme en Ruby (<<), qui modifie la chaine sans en créer une nouvelle?

Oui, le + crée une nouvelle chaine, faudrait utiliser un buffer pour faire du append, mais dans le cas là je vois pas trop l'interêt c'est pas comme si je le faisais souvent !

tremblay-guy · 2017-03-17T12:36:39Z

+
+	for c in creatures do c.async.run
+
+	active_actors.is_empty


Que fait "is_empty"? Avec un tel nom, ça devrait être un prédicat, donc une méthode qui retourne un booléen... ce qui ne semble pas approprié ici puisque le résultat n'est pas utilisé.

Donc ça doit être une commande... mais avec un drôle de nom

HAHA c'est une particularité de ma ReverseBlockingQueue, il faut que je la change mais je ferais ça plus tard, enfait le is_empty est bloquant sur cette liste tant qu'il reste des éléments dedans. C'est mon point de synchronisation pour savoir quand mes acteurs ont plus de message a traiter !

tremblay-guy · 2017-03-17T12:37:52Z

+
+	# Receive a token, then send it to `next` unless its value is `0`
+	fun send_token(message: Int) do
+		if message == 0 then


if message == 0 then print id # Une seule ligne?

if message >= 1 then next.async... # Idem

Ah oui, sûrement des restes de quand il y avait du code de debug a cet endroit !

tremblay-guy · 2017-03-17T12:38:17Z

+var first = new ThreadRing(1)
+var current = new ThreadRing(2)
+first.next = current
+for i in [3..503] do


Constante magique 503 => Constante symbolique!!

@tremblay-guy

Tried to follow @tremblay-guy commentaries on #2361 to improve the code of the actors examples Pull-Request: #2401 Reviewed-by: Alexis Laferrière <alexis.laf@xymus.net>

BlackMinou added the need_review label Feb 6, 2017

privat approved these changes Feb 8, 2017

View reviewed changes

BlackMinou force-pushed the actors branch from 376e890 to a79cf54 Compare February 15, 2017 20:07

xymus reviewed Feb 16, 2017

View reviewed changes

BlackMinou added 4 commits February 20, 2017 20:23

Actors abstraction

c092407

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Adding a Nitty version of a condition variable

55f96ee

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

thread ring example from benchmarksgame

cc7bbf7

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Adds an atomic Int to pthreads

3f8d050

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

BlackMinou force-pushed the actors branch from a79cf54 to 6cb5fa1 Compare February 21, 2017 01:30

BlackMinou added 4 commits February 20, 2017 20:32

Mandelbrot example from benchmarks game

e2f2e01

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Chameneos-redux actor example

50d12c8

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Fannkuch-redux actor example

7f7c2a5

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Actors model injection

e5c9f15

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

BlackMinou force-pushed the actors branch from 6cb5fa1 to 706b52f Compare February 21, 2017 01:33

xymus approved these changes Feb 21, 2017

View reviewed changes

BlackMinou added 8 commits February 27, 2017 10:06

Actors support module generation phase

9baf4cf

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Adding a hook on a module at the end of a phase

4926589

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Adding actor phases to the frontend

ceebe1a

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

Adding a README to the actors group

6e979d5

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

A very simple example using actors

15dd7c6

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

initialize $WRITE for tests others than *.args

4a42952

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

ignoring actors_*.nit files generated by nitc

bb4daab

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

skip the execution of tests on actors_*.nit files

17cd19a

Signed-off-by: BlackMinou <romain.chanoir@viacesi.fr>

BlackMinou force-pushed the actors branch from d6ce072 to 17cd19a Compare February 27, 2017 20:06

BlackMinou mentioned this pull request Feb 27, 2017

Fictive module in memory instead of on the disk #2378

Open

privat merged commit 17cd19a into nitlang:master Feb 28, 2017

privat removed the need_review label Feb 28, 2017

tremblay-guy reviewed Mar 17, 2017

View reviewed changes

BlackMinou mentioned this pull request Mar 29, 2017

Some minor fixes to actors examples #2401

Merged

privat added a commit that referenced this pull request Apr 5, 2017

Merge: Some minor fixes to actors examples

5bf1ff8

Tried to follow @tremblay-guy commentaries on #2361 to improve the code of the actors examples Pull-Request: #2401 Reviewed-by: Alexis Laferrière <alexis.laf@xymus.net>

		fun wait(mutex: NativePthreadMutex) `{ pthread_cond_wait(self, mutex); `}
		end


		## Waiting for all actors to finish processing

		Let's imagine you create a whole bunch of actors and make them do things asynchronously from the main thread. Your don't want your program to exit right after giving work to your actors.


		var module_path = "{mmodule_path}/{module_name}.nit"

		var nit_module = new NitModule(module_name + " is no_warning(\"missing-doc\")")

		print chk.to_s + "\nPfannfuchen(" + n.to_s + ") = " + res.to_s

		end

Conversation

BlackMinou commented Feb 6, 2017

Uh oh!

privat left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

BlackMinou commented Feb 15, 2017

Uh oh!

xymus commented Feb 16, 2017

Uh oh!

xymus left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

xymus commented Feb 16, 2017

Uh oh!

xymus left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tremblay-guy Mar 17, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tremblay-guy Mar 17, 2017 •

edited

Loading

tremblay-guy Mar 17, 2017 •

edited

Loading

tremblay-guy Mar 17, 2017 •

edited

Loading