Levente, I never said that. I said that if it were built properly and used naively, it would have led to problems. Enter timeouts that should not be present. The server side is almost impossible to defend as it stands.
Sockets don't decide when to free resources. Threads and users do. Bill ________________________________________ From: [email protected] [[email protected]] On Behalf Of Levente Uzonyi [[email protected]] Sent: Thursday, September 23, 2010 12:25 PM To: [email protected] Subject: Re: [Pharo-project] we need to find a way to declare that an action is UIless On Thu, 23 Sep 2010, Schwab,Wilhelm K wrote: > Levente, > > We can twist words indefinitely. I have been describing a blocking connect, > because that is precisely what one should be trying to do: put one thread on > hold until the calling thread is connected. There is no sensible default > waiting period for that to happen, and so the framework should not be asking > for a time limit at all, let alone insisting on one. My example was just a proof against the "we need a new socket implementation because the current one blocks the image" theory. And you're wrong about the timeouts. If sockets could wait indefinitely, the chance for resource leakage would be very high. I'm pretty sure that even if you omit the timeout (which is possible with the current API) there will be another timeout at the OS level which you can't/shouldn't work around. Here is an example for a blocking connection without a timeout: Transcript open. [ | s | [ s := Socket new. s connectNonBlockingTo: #[172 16 0 1] port: 12345. Transcript show: 'Connecting...'; cr. [ s statusString = 'waitingForConnection' ] whileTrue: [ s semaphore wait. "No timeout." ]. Transcript show: s statusString; cr. s close ] ensure: [ s destroy ] ] fork. > > ConnectionQueue being at the heart of a Squeak socket server is not my idea; > I read that in various places, tried, and was appalled to discover that it > times out, returns nil, etc. It is a complete mess that polls for a time > period when it should be blocking a thread until an event occurs. ConnectionQueue is just a high level API, you can always use _Sockets_. Levente > > Bill > > ________________________________________ > From: [email protected] > [[email protected]] On Behalf Of Levente Uzonyi > [[email protected]] > Sent: Wednesday, September 22, 2010 11:51 PM > To: [email protected] > Subject: Re: [Pharo-project] we need to find a way to declare that an action > is UIless > > On Wed, 22 Sep 2010, Schwab,Wilhelm K wrote: > >> Levente, >> >> Something has to block while a connection attempt is pending, and not just >> until some arbitrary time limit. The client code is bad enough; the servers >> are horrible. > > If the UI Process is using the Socket and it's using a blocking connection > method, then - no surprise - it will be blocked. This won't affect other > processes. > > If you were right, the following would block the UI Process for 100 > seconds, but it doesn't block it at all, just try it: > > Transcript open. > [ 10 timesRepeat: [ > | s | > [ > s := Socket new. > s connectNonBlockingTo: #[172 16 0 1] port: 12345. > s > waitForConnectionFor: 10 > ifTimedOut: [ Transcript show: 'Couldn''t connect.'; > cr ]. > s isConnected ifTrue: [ > Transcript show: 'Connected.'; cr. ]. > s close ] ensure: [ > s ifNotNil: [ s destroy ] ] ] ] fork. > > > What do you mean by "servers"? ConnectionQueue? > > > Levente > >> >> Bill >> >> >> ________________________________________ >> From: [email protected] >> [[email protected]] On Behalf Of Levente Uzonyi >> [[email protected]] >> Sent: Wednesday, September 22, 2010 9:55 PM >> To: [email protected] >> Subject: Re: [Pharo-project] we need to find a way to declare that an action >> is UIless >> >> On Mon, 20 Sep 2010, Schwab,Wilhelm K wrote: >> >>> the one running the gui >> >> In that case, you're wrong. The UI Process will be able to run, because >> other processes using Sockets will wait on Semaphores and not because of >> "time limits". So I just convinced myself (and hopefully you too) about >> that using Socket instances will not hang the entire image, just the >> Process that uses the Socket. Therefore the SocketPlugin is as good as >> possible. >> >> >> Levente >> >>> >>> >>> >>> >>> ________________________________________ >>> From: [email protected] >>> [[email protected]] On Behalf Of Levente Uzonyi >>> [[email protected]] >>> Sent: Monday, September 20, 2010 10:14 PM >>> To: [email protected] >>> Subject: Re: [Pharo-project] we need to find a way to declare that an >>> action is UIless >>> >>> On Mon, 20 Sep 2010, Schwab,Wilhelm K wrote: >>> >>>> Scratch around, and you will find that the time limits are there to allow >>>> calls to made on the main thread. >>> >>> Where? In the Socket class? And what's the "main thread"? >>> >>> >>> Levente >>> >>>> >>>> >>>> >>>> ________________________________________ >>>> From: [email protected] >>>> [[email protected]] On Behalf Of Levente Uzonyi >>>> [[email protected]] >>>> Sent: Monday, September 20, 2010 8:13 PM >>>> To: [email protected] >>>> Subject: Re: [Pharo-project] we need to find a way to declare that an >>>> action is UIless >>>> >>>> On Mon, 20 Sep 2010, Schwab,Wilhelm K wrote: >>>> >>>>> Levente, >>>>> >>>>> If they worked correctly, they would, at least under naive client >>>>> conditions - a connection attempt should try until it is told >>>>> (#terminate) to stop. Severs that listen for a limited time are broken >>>>> by design. ConnectionQueue polls as a result - it's pretty bad. >>>> >>>> I guess you're using Socket>>#connectTo:port: which uses >>>> Socket class>>#standardTimeout as timeout (45 seconds). If you don't want >>>> the default timeout, use #connectTo:port:waitForConnectionFor: or >>>> implement your own low level method which waits on semaphore until it's >>>> signaled. If you want to terminate the connection attempt, just signal the >>>> semaphore yourself, like here: >>>> >>>> s := Socket newTCP. >>>> s connectNonBlockingTo: #[127 0 0 1] port: 19327. "Random port which is >>>> not open." >>>> [ 500 milliSeconds asDelay wait. s semaphore signal ] fork. "This process >>>> will stop the connection attempt." >>>> s semaphore waitTimeoutMSecs: 1000. >>>> s statusString. "This will simply print the socket status. You can >>>> terminate the process here if the socket is connected, etc." >>>> >>>> And for ConnectionQueue, simply don't use it if you don't like it. It >>>> doesn't have to poll at all, AFAIK it's just implemented that way. >>>> Since Sockets use Semaphores which are signaled by the SocketPlugin, they >>>> don't block the image at all. But correct me if I'm wrong. >>>> >>>> >>>> Levente >>>> >>>>> >>>>> Bill >>>>> >>>>> >>>>> >>>>> ________________________________________ >>>>> From: [email protected] >>>>> [[email protected]] On Behalf Of Levente Uzonyi >>>>> [[email protected]] >>>>> Sent: Monday, September 20, 2010 7:01 PM >>>>> To: [email protected] >>>>> Subject: Re: [Pharo-project] we need to find a way to declare that an >>>>> action is UIless >>>>> >>>>> On Mon, 20 Sep 2010, Schwab,Wilhelm K wrote: >>>>> >>>>>> Guillermo, >>>>>> >>>>>> One has to be careful with assuming that something affects only part of >>>>>> the image. Much of Squeak's networking trouble comes from the fact that >>>>>> it was designed to block the image for a limited time when it should >>>>>> have been blocking only one Process *indefinitely*. But, the remedy for >>>>>> working while blocking operations happen in the background is threading, >>>>>> and most of the image is deliberately not thread safe. >>>>> >>>>> When does a Socket block the image? >>>>> >>>>> >>>>> Levente >>>>> >>>>>> >>>>>> Bill >>>>>> >>>>>> >>>>>> ________________________________________ >>>>>> From: [email protected] >>>>>> [[email protected]] On Behalf Of Guillermo >>>>>> Polito [[email protected]] >>>>>> Sent: Sunday, September 19, 2010 10:56 PM >>>>>> To: [email protected] >>>>>> Subject: Re: [Pharo-project] we need to find a way to declare that an >>>>>> action is UIless >>>>>> >>>>>> +1 to Bill's. If we can't have a feedback from the system while doing >>>>>> silent actions, we can think it just freezed :S. >>>>>> >>>>>> And it's something already dicussed, but I don't like actions that >>>>>> affect only a part of the system blocking my whole image. >>>>>> >>>>>> Guille >>>>>> >>>>>> On Sun, Sep 19, 2010 at 10:40 PM, Igor Stasenko >>>>>> <[email protected]<mailto:[email protected]>> wrote: >>>>>> On 20 September 2010 03:09, Schwab,Wilhelm K >>>>>> <[email protected]<mailto:[email protected]>> wrote: >>>>>>> Slow access can be a big problem. Any such change should be made based >>>>>>> on measurements so we know what benefit we get at what cost. >>>>>>> >>>>>> Yeah, it would be much easier to deal that line in Self or JavaScript, >>>>>> where you can add any properties to object >>>>>> on the fly, without need of adding a methods or declaring additional >>>>>> instance variable in class... >>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> ________________________________________ >>>>>>> From: >>>>>>> [email protected]<mailto:[email protected]> >>>>>>> >>>>>>> [[email protected]<mailto:[email protected]>] >>>>>>> On Behalf Of Igor Stasenko >>>>>>> [[email protected]<mailto:[email protected]>] >>>>>>> Sent: Sunday, September 19, 2010 7:56 PM >>>>>>> To: >>>>>>> [email protected]<mailto:[email protected]> >>>>>>> Subject: Re: [Pharo-project] we need to find a way to declare that an >>>>>>> action is UIless >>>>>>> >>>>>>> On 19 September 2010 13:12, Stéphane Ducasse >>>>>>> <[email protected]<mailto:[email protected]>> wrote: >>>>>>>> hi guys >>>>>>>> >>>>>>>> I tried to add borderStyle to BorderedMorph and the progressbar >>>>>>>> showing the progress blow up. >>>>>>>> So we should really have a way to specify silent ui action. >>>>>>>> Does anybody have an idea how I could do that? >>>>>>>> >>>>>>>> [BorderedMorph addInstVarNamed: 'borderStyle'] silent would be >>>>>>>> cool. >>>>>>>> >>>>>>> >>>>>>> use morph propertyAt: #borderStyle >>>>>>> so you don't have to break your head with it :) >>>>>>> >>>>>>> BorderedMorph having an enormous number of subclasses, while some of >>>>>>> them even don't using any >>>>>>> kind of borders. That's makes me wonder if anything like color, border >>>>>>> style etc should belong to root classes >>>>>>> in hierarchy, like Morph or BorderedMorph. I think that dynamic set of >>>>>>> properties (which is currently sits in morphic >>>>>>> extensions are more appropriate storage for them). The only problem is >>>>>>> that accessing them is much slower than ivars. >>>>>>> >>>>>>>> Stef >>>>>>>> _______________________________________________ >>>>>>>> Pharo-project mailing list >>>>>>>> [email protected]<mailto:[email protected]> >>>>>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> -- >>>>>>> Best regards, >>>>>>> Igor Stasenko AKA sig. >>>>>>> >>>>>>> _______________________________________________ >>>>>>> Pharo-project mailing list >>>>>>> [email protected]<mailto:[email protected]> >>>>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>>>>>> >>>>>>> _______________________________________________ >>>>>>> Pharo-project mailing list >>>>>>> [email protected]<mailto:[email protected]> >>>>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>>>>>> >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> Best regards, >>>>>> Igor Stasenko AKA sig. >>>>>> >>>>>> _______________________________________________ >>>>>> Pharo-project mailing list >>>>>> [email protected]<mailto:[email protected]> >>>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>>>>> >>>>>> >>>>>> _______________________________________________ >>>>>> Pharo-project mailing list >>>>>> [email protected] >>>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>>>>> >>>>> >>>>> _______________________________________________ >>>>> Pharo-project mailing list >>>>> [email protected] >>>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>>>> >>>> >>>> _______________________________________________ >>>> Pharo-project mailing list >>>> [email protected] >>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>>> >>> >>> _______________________________________________ >>> Pharo-project mailing list >>> [email protected] >>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >>> >> >> _______________________________________________ >> Pharo-project mailing list >> [email protected] >> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project >> > > _______________________________________________ > Pharo-project mailing list > [email protected] > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > _______________________________________________ Pharo-project mailing list [email protected] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
