www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.announce - phobosx.signal ready

reply "Robert" <jfanatiker gmx.at> writes:
phobosx.signal ready

I finished my std.signals2 implementation. I moved it to:

https://github.com/phobos-x/phobosx/blob/master/source/phobosx/signal.d

you can also find it in the dub registry:

http://code.dlang.org/packages/phobosx

Documentation:

https://vhios.dyndns.org/signal.html

I renamed it from std.signals2 to phobosx.signal. When included in
phobos it will most likely be named std.signal.

Improvements over std.signals:

- No more heisenbugs 
(http://d.puremagic.com/issues/show_bug.cgi?id=4150)

- safe connect method to an object's method (weak ref semantic)

- support for indirect connect to an object's method via a wrapper
   delegate (weak ref semantics to target object)

- strongConnect method for connecting to non objects. (strong ref
   semantics)

- disconnect method for disconnecting all methods of a given 
object with
   a single call.

- Proper exception handling. Meaning if a slot throws an 
exception the
   other slots will still be called and all slot exceptions are 
chained
   together.

- signal methods are safe to call from within a slot.

- signal copying is forbidden. While this implementation would not
   cause segfaults when copied like std.signals, it is not really 
clear
   how a signal should be copied in the general case.

- Signals are implemented as structs instead of a mixin and every
   template parameter agnostic code also does not depend on the 
template
   parameters -> avoid template bloat.

- Memory footprint for empty signal is even lower than the one of 
the
   current std.signals, which is important because many signals 
end up not
   being used at all.

- Provided string mixin which allows only the containing class to
   issue emit(), but everyone to connect.

- Slots are called in the same sequence as they were registered.

I could not find any more bugs, so please use it and show me the 
bugs I missed :-)

Have fun!

Best regards,

Robert
Jul 15 2013
next sibling parent reply David <d dav1d.de> writes:
Thanks for your work, I'll try to test it the next days!
Jul 17 2013
parent "Robert" <jfanatiker gmx.at> writes:
On Wednesday, 17 July 2013 at 16:06:14 UTC, David wrote:

 Thanks for your work, I'll try to test it the next days!


Thanks! I really appreciate it. The more real world use, the 
better and the sooner it can be moved to phobos! I am going to 
fix any found issues ASAP.
Jul 18 2013
prev sibling next sibling parent reply "Zoadian" <github zoadian.de> writes:
Thank you very much!
i might finally get rid of all those segfaults now :)
Jul 17 2013
next sibling parent reply "Zoadian" <github zoadian.de> writes:
you might run into 
http://d.puremagic.com/issues/show_bug.cgi?id=10260

drop the -profile switch to make it work
Jul 17 2013
parent reply "Zoadian" <github zoadian.de> writes:
I'm also running into 
http://d.puremagic.com/issues/show_bug.cgi?id=10502 when using 
your mixin signal function.

So I'm using Signal! directly now.
Works well so far.


Btw this forum needs an edit button!
Jul 17 2013
next sibling parent "John Colvin" <john.loughran.colvin gmail.com> writes:
On Wednesday, 17 July 2013 at 19:38:54 UTC, Zoadian wrote:

 Btw this forum needs an edit button!


Not gonna happen as it's not a forum, it's just a forum interface 
to an NNTP newsgroup and mailing list.
Jul 17 2013
prev sibling parent "Robert" <jfanatiker gmx.at> writes:
On Wednesday, 17 July 2013 at 19:38:54 UTC, Zoadian wrote:

 I'm also running into 
 http://d.puremagic.com/issues/show_bug.cgi?id=10502 when using 
 your mixin signal function.

 So I'm using Signal! directly now.
 Works well so far.


 Btw this forum needs an edit button!


Thanks for pointing out those issues! I added a link to them to 
the documentation for now. This does not fix anything of course, 
but at least one does not need to find out the hard way ;-)
Jul 18 2013
prev sibling parent "Robert" <jfanatiker gmx.at> writes:
On Wednesday, 17 July 2013 at 17:23:48 UTC, Zoadian wrote:

 Thank you very much!
 i might finally get rid of all those segfaults now :)


You are very welcome. I am glad to see some feedback, I was 
already afraid I did all the work and nobody uses it. I myself 
currently have no use case for signals in D, but I used them a 
lot in C++ and found the D implementation not quite satisfying, 
so I had to improve it.

Best regards,

Robert
Jul 18 2013
prev sibling next sibling parent reply "E.S. Quinn" <anonymous example.com> writes:
Out of curiosity, does this implementation support e.g. function 
literals or nested functions with connect()? The current phobos 
implementation only allows connecting class methods, and that's a 
limitation I've bumped up against quite a bit, especially when 
writing test code.
Jul 17 2013
parent reply "Robert" <jfanatiker gmx.at> writes:
On Thursday, 18 July 2013 at 01:26:17 UTC, E.S. Quinn wrote:

 Out of curiosity, does this implementation support e.g. 
 function literals or nested functions with connect()? The 
 current phobos implementation only allows connecting class 
 methods, and that's a limitation I've bumped up against quite a 
 bit, especially when writing test code.


I just improved the example given in the docs, see the code after 
comment '//Do some fancy stuff'. It should answer your question. 
;-)

https://vhios.dyndns.org/signal.html

Best regards,

Robert
Jul 18 2013
parent reply "John Colvin" <john.loughran.colvin gmail.com> writes:
On Thursday, 18 July 2013 at 08:44:39 UTC, Robert wrote:

 On Thursday, 18 July 2013 at 01:26:17 UTC, E.S. Quinn wrote:

 Out of curiosity, does this implementation support e.g. 
 function literals or nested functions with connect()? The 
 current phobos implementation only allows connecting class 
 methods, and that's a limitation I've bumped up against quite 
 a bit, especially when writing test code.


 I just improved the example given in the docs, see the code 
 after comment '//Do some fancy stuff'. It should answer your 
 question. ;-)

 https://vhios.dyndns.org/signal.html

 Best regards,

 Robert


Your site gives me an ssl certificate error with https and a 504 
with http
Jul 18 2013
parent reply "Robert" <jfanatiker gmx.at> writes:
 Your site gives me an ssl certificate error with https and a 
 504 with http


Yeah :-) It is only a home server, I did not pay for a proper 
certificate. Just ignore the security warning. I merely wanted to 
publish the documentation somewhere for easier access.

Best regards,

Robert
Jul 18 2013
parent reply "Brad Anderson" <eco gnuk.net> writes:
On Thursday, 18 July 2013 at 11:31:06 UTC, Robert wrote:

 Your site gives me an ssl certificate error with https and a 
 504 with http


 Yeah :-) It is only a home server, I did not pay for a proper 
 certificate. Just ignore the security warning. I merely wanted 
 to publish the documentation somewhere for easier access.

 Best regards,

 Robert


You can get free certificates from StartSSL.
Jul 23 2013
parent "Robert" <jfanatiker gmx.at> writes:
 You can get free certificates from StartSSL.


Thanks, that's actually cool :-) Unfortunately it does not work 
for my dyndns domain. (I can't register vhios.dyndns.org, only 
dyndns.org would be accepted)
Jul 24 2013
prev sibling parent reply David <d dav1d.de> writes:
Ok, I came around to test it, finally.


A few things:
 * I don't like the separation between connect/strongConnect
 * I would like to bind a strongConnect to a class, if the class is gone
the strongConnect should be gone, too. E.g.
"window.single_key_down[JUMP].strongConnect({ physics.jump(); });"
physics is a class variable, this will blow up once the class holding it
is gone
 * I don't like the connect api for class-method connects, I write all
the time "window.on_mouse_pos.connect!"on_mouse_pos"(this);" (the this
is kinda annoying), doesn't something like signal.connect!(bla)() work,
with an alias template param?
 * Why is there no convenience class for Signal? Since the copy
constructor is disabled, I have to use a pointer, not a big deal, but a
class would be more convenient.


Other than that! Great job, I had not a single segfault so far (only 3
"starts" so far though :>)
Jul 21 2013
next sibling parent "Robert" <jfanatiker gmx.at> writes:
 A few things:
  * I don't like the separation between connect/strongConnect


It should be possible to make strongConnect just another overload 
of connect, but I chose to use another name because of the 
changed semantics. This way both the writer and the reader of the 
code have a clue what is going on. If strongConnect was also 
called connect, you could easily choose a strong connection by 
accident: sig.connect(&o.watch); // Oops strong connect.


  * I would like to bind a strongConnect to a class, if the 
 class is gone
 the strongConnect should be gone, too. E.g.
 "window.single_key_down[JUMP].strongConnect({ physics.jump(); 
 });"
 physics is a class variable, this will blow up once the class 
 holding it
 is gone


If physics is on the GC collected heap - nothing is going to blow 
up. The signal just keeps a strong ref to physics so the GC won't 
free it. If physics is for example reference counted or its 
lifetime is managed manually then it would blow up if you forget 
to call disconnect before destroying the object.

It would be easier if you used connect instead of strongConnect, 
this way the connection is dropped whenever physics gets 
destroyed, whatever the cause is. Also the signal won't keep it 
alive.

If you really mean a class by 'class variable' than it can't 
really be gone, except maybe for dynamic linking. In this case 
you would have to manage your connections yourself.


  * I don't like the connect api for class-method connects, I 
 write all
 the time "window.on_mouse_pos.connect!"on_mouse_pos"(this);" 
 (the this
 is kinda annoying), doesn't something like 
 signal.connect!(bla)() work,
 with an alias template param?


Unfortunately I don't see how this could work, but I am open for 
suggestions.


  * Why is there no convenience class for Signal? Since the copy
 constructor is disabled, I have to use a pointer, not a big 
 deal, but a
 class would be more convenient.


Not sure about this. It would just be a wrapper doing nothing 
except forwarding function calls, is this really better:

  auto sig = new SignalWrapper!(int)();
  void func(SignalWrapper!int sig) {}

than

  auto sig = new Signal!(int)();
  void func(Signal!int* sig) {}

?


 Other than that! Great job, I had not a single segfault so far 
 (only 3
 "starts" so far though :>)


:-)
Jul 24 2013
prev sibling parent reply "Kagamin" <spam here.lot> writes:
On Sunday, 21 July 2013 at 11:24:42 UTC, David wrote:

 "window.single_key_down[JUMP].strongConnect({ physics.jump(); 
 });"
 physics is a class variable, this will blow up once the class 
 holding it
 is gone


You ask for the weak ref semantics, but as the name suggests, 
strongConnect has the strong ref semantics. In this case the 
closure keep a strong ref to the object with the physics field 
and the signal keeps a strong ref to the closure, so the window 
keeps a strong ref to the object, so it won't be collected.
Jul 28 2013
parent David <d dav1d.de> writes:
Am 28.07.2013 18:39, schrieb Kagamin:

 On Sunday, 21 July 2013 at 11:24:42 UTC, David wrote:

 "window.single_key_down[JUMP].strongConnect({ physics.jump(); });"
 physics is a class variable, this will blow up once the class holding it
 is gone


 
 You ask for the weak ref semantics, but as the name suggests,
 strongConnect has the strong ref semantics. In this case the closure
 keep a strong ref to the object with the physics field and the signal
 keeps a strong ref to the closure, so the window keeps a strong ref to
 the object, so it won't be collected.


Yeah I noticed that shortly after I wrote that... the "connect" method
should just work.
Jul 28 2013