Pharo windows hanging ...
Locked
 
|
Igor Stasenko wrote:
> This is because all do-its executed in process, which responsible from > handling UI events. > To avoid 'hanging' you can just do like following: > [ do what you need ] fork. Oh, yes! Then i commanded a Full GC. I had to clean some 16 MByte Smalltalk stuff on my Notebook. 15 Seconds I had to wait, till my keystrokes and mouse events were accepted back again by Pharo GUI. Well, in the meantime, the seaside process was unresponsive. Ok, no real problem. Customers can wait and I have my coffee breaks ;-) World Wide Wait with Seaside Servers on Pharo ... Sorry, no access to The Salty Spitoon .... (not yet) ;-) Have fun, Guido Stepken _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
2009/4/22 stepken <[hidden email]>:
> Igor Stasenko wrote: > > > This is because all do-its executed in process, which responsible from > > handling UI events. > > To avoid 'hanging' you can just do like following: > > [ do what you need ] fork. > > Oh, yes! > > Then i commanded a Full GC. I had to clean some 16 MByte Smalltalk stuff > on my Notebook. 15 Seconds I had to wait, till my keystrokes and mouse > events were accepted back again by Pharo GUI. > > Well, in the meantime, the seaside process was unresponsive. > > Ok, no real problem. Customers can wait and I have my coffee breaks ;-) > > World Wide Wait with Seaside Servers on Pharo ... > threading model which means that only a single process can be active at a single point of time, and the order how they proceed are determined by scheduler. I understand the reasons of your sarcasm, yes.. its not perfect! Many, many code in Squeak (and consequently in Pharo) is naively implemented (including UI), and without concurrency in mind. But this is not the only UI framework which hates the concurrency - take a look at "groundbreaking" Mac OS :) > Sorry, no access to The Salty Spitoon .... (not yet) ;-) > > Have fun, Guido Stepken > > > > _______________________________________________ > Pharo-project mailing list > [hidden email] > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > -- Best regards, Igor Stasenko AKA sig. _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
In reply to this post by stepken
Hi, Igor!
I've got a *brand new* design idea for PHARO ;-) All events (keyboard, mouse, network etc.) should be read by a separate process, which i call ... hmmmm ... "event dispatcher". Brilliant name, isn't it? That one reads the event from the operating system, puts it into a per-windowgroup "event-queue", and notifies the "view process" or "network process" about the arrival of the event. That one sits on a semaphore, waiting for the arrival. regards, Guido Stepken _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
2009/4/23 stepken <[hidden email]>:
> Hi, Igor! > > I've got a *brand new* design idea for PHARO ;-) > > All events (keyboard, mouse, network etc.) should be read by a separate > process, which i call ... hmmmm ... "event dispatcher". Brilliant name, > isn't it? > :) > That one reads the event from the operating system, puts it into a > per-windowgroup "event-queue", and notifies the "view process" or > "network process" about the arrival of the event. That one sits on a > semaphore, waiting for the arrival. > Unless you going to write own *brand new* UI framework. Because modifying existing one leads to madness & frustration.. - ask Stephane ;) And in fact, things are more or less similar to what you described: there is a separate process which fetching events from VM, and then queuing them to be handled by UI process. > regards, Guido Stepken > > > > > > _______________________________________________ > Pharo-project mailing list > [hidden email] > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > -- Best regards, Igor Stasenko AKA sig. _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
In reply to this post by Igor Stasenko
On Apr 22, 2009, at 11:38 AM, Igor Stasenko wrote: But this is not the only UI framework which hates the concurrency - yes, but they have a nice mechanism for dealing with this when a worker thread wants to update the UI - (void)performSelectorOnMainThread:(SEL)aSelector withObject:(id)arg waitUntilDone:(BOOL)wait; Is there a Squeak equivalent for getting something invoked on the main UI thread? -Todd Blanchard _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
On Wed, Apr 22, 2009 at 09:17:12PM -0700, Eagle Offshore wrote:
> > On Apr 22, 2009, at 11:38 AM, Igor Stasenko wrote: > > >But this is not the only UI framework which hates the concurrency - > >take a look at "groundbreaking" Mac OS :) > > yes, but they have a nice mechanism for dealing with this when a > worker thread wants to update the UI > > - (void)performSelectorOnMainThread:(SEL)aSelector withObject:(id)arg > waitUntilDone:(BOOL)wait; > > Is there a Squeak equivalent for getting something invoked on the main > UI thread? WorldState class>>addDeferredUIMessage: _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
In reply to this post by Eagle Offshore
Ok, I have to comment since i'm up to my neck with alligators over
performSelectorOnMainThread: In fact just this evening 4 hours toasted to document a bug with this call with Apple for tomorrow. (a) people don't realize with the carbon os-x vm we run on the Main Thread and poll for os-x events that then get dispatch to the squeak vm event handlers. This leads to the interesting jerky non-responsive feedback you get with the VM when you accidently kill the morphic polling loop, since we aren't servicing the UI event polling fast enough since we are using a fallback 1/2 second ancient piece of code that in the past looked for the interrupt key. (b) If you use the Unix os-x vm, that spins the squeak thread off as a background task, now if you do UI related work via FFI then you die. (c) performSelectorOnMainThread: has lots of interesting bugs on the iPhone, especially if you try to do view animation with it. (d) the iphone vm runs as a background thread, but deadlocks with UIWebView between it and the main thread, are interesting, all solvable, but interesting. Fortunately in the future we can migrate back to (a) on the iPhone. (e) if you run on the Main Thread then you need something to handle event callbacks from the UI when you call the UI to service a UI event. Fortunately this has been worked out and is easier than (d) Frankly (a), (b), (c) and did I mention (d) are serious complications I could do without. On 22-Apr-09, at 9:17 PM, Eagle Offshore wrote: > > On Apr 22, 2009, at 11:38 AM, Igor Stasenko wrote: > >> But this is not the only UI framework which hates the concurrency - >> take a look at "groundbreaking" Mac OS :) > > yes, but they have a nice mechanism for dealing with this when a > worker thread wants to update the UI > > - (void)performSelectorOnMainThread:(SEL)aSelector withObject: > (id)arg waitUntilDone:(BOOL)wait; > > Is there a Squeak equivalent for getting something invoked on the > main UI thread? > > -Todd Blanchard > _______________________________________________ > Pharo-project mailing list > [hidden email] > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project -- = = = ======================================================================== John M. McIntosh <[hidden email]> Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com = = = ======================================================================== _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
Re: Pharo windows hanging ...
|
|
John M McIntosh wrote:
> Ok, I have to comment since i'm up to my neck with alligators over > performSelectorOnMainThread: In fact just this evening 4 hours toasted > to document a bug with this call with Apple for tomorrow. > > (a) people don't realize with the carbon os-x vm we run on the Main > Thread and poll for os-x events that then get dispatch to the squeak > vm event handlers. > This leads to the interesting jerky non-responsive feedback you get > with the VM when you accidently kill the morphic polling loop, since > we aren't servicing > the UI event polling fast enough since we are using a fallback 1/2 > second ancient piece of code that in the past looked for the > interrupt key. This is gone in the new event sensor design. Michael _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
In reply to this post by johnmci
I assume c and d will get fixed (since doing animation from worker
threads is pretty much the point). b - well of course you can't do UI work from FFI - you're not on the UI thread - unless you call performSelectorOnMainThread:... from there. a - I dunno - it seems like squeak's vm ought to do most work not on the UI thread except for UI things but the runtime models aren't exactly what I'd call fundamentally compatible since Cocoa wants to own the runloop and just call you back now and then via delegates or notifications. I have a Cocoa MIDI/CoreAudio AudioUnits host that makes heavy use of coreaudio (which ends up spawning a few dozen threads) but still animates pictures of keyboards, sound meters, etc.... in near real time and does it all by calling performSelectorOnMainThread:... and it all works like a charm so long as I remember to invoke the graphics update routines via performSelectorOnMainThread:. I would like to try out the new Cocoa bridge on a Mac at some point and see what can be done with it. -Todd Blanchard On Apr 22, 2009, at 10:30 PM, John M McIntosh wrote: > Ok, I have to comment since i'm up to my neck with alligators over > performSelectorOnMainThread: In fact just this evening 4 hours toasted > to document a bug with this call with Apple for tomorrow. > > (a) people don't realize with the carbon os-x vm we run on the Main > Thread and poll for os-x events that then get dispatch to the squeak > vm event handlers. > This leads to the interesting jerky non-responsive feedback you get > with the VM when you accidently kill the morphic polling loop, since > we aren't servicing > the UI event polling fast enough since we are using a fallback 1/2 > second ancient piece of code that in the past looked for the > interrupt key. > > (b) If you use the Unix os-x vm, that spins the squeak thread off as a > background task, now if you do UI related work via FFI then you die. > > (c) performSelectorOnMainThread: has lots of interesting bugs on the > iPhone, especially if you try to do view animation with it. > > (d) the iphone vm runs as a background thread, but deadlocks with > UIWebView between it and the main thread, are interesting, all > solvable, but interesting. Fortunately in the future we can migrate > back to (a) on the iPhone. > > (e) if you run on the Main Thread then you need something to handle > event callbacks from the UI when you call the UI to service a UI > event. Fortunately this has been worked out and is easier than (d) > > Frankly (a), (b), (c) and did I mention (d) are serious complications > I could do without. > > On 22-Apr-09, at 9:17 PM, Eagle Offshore wrote: > >> >> On Apr 22, 2009, at 11:38 AM, Igor Stasenko wrote: >> >>> But this is not the only UI framework which hates the concurrency - >>> take a look at "groundbreaking" Mac OS :) >> >> yes, but they have a nice mechanism for dealing with this when a >> worker thread wants to update the UI >> >> - (void)performSelectorOnMainThread:(SEL)aSelector withObject: >> (id)arg waitUntilDone:(BOOL)wait; >> >> Is there a Squeak equivalent for getting something invoked on the >> main UI thread? >> >> -Todd Blanchard >> _______________________________________________ >> Pharo-project mailing list >> [hidden email] >> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > > -- > = > = > = > = > = > ====================================================================== > John M. McIntosh <[hidden email]> > Corporate Smalltalk Consulting Ltd. http:// > www.smalltalkconsulting.com > = > = > = > = > = > ====================================================================== > > > > > _______________________________________________ > Pharo-project mailing list > [hidden email] > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
Re: Pharo windows hanging ...
|
|
In reply to this post by stepken
You are not forced to use pharo and you can send code once you signed
the license agreement. talking is easier than doing. Stef (a Doer). > Igor Stasenko wrote: > >> This is because all do-its executed in process, which responsible >> from >> handling UI events. >> To avoid 'hanging' you can just do like following: >> [ do what you need ] fork. > > Oh, yes! > > Then i commanded a Full GC. I had to clean some 16 MByte Smalltalk > stuff > on my Notebook. 15 Seconds I had to wait, till my keystrokes and mouse > events were accepted back again by Pharo GUI. > > Well, in the meantime, the seaside process was unresponsive. > > Ok, no real problem. Customers can wait and I have my coffee > breaks ;-) > > World Wide Wait with Seaside Servers on Pharo ... > > Sorry, no access to The Salty Spitoon .... (not yet) ;-) > > Have fun, Guido Stepken > > > > _______________________________________________ > Pharo-project mailing list > [hidden email] > http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project > _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
|
|
In reply to this post by Eagle Offshore
On 23-Apr-09, at 12:40 AM, Eagle Offshore wrote: > I assume c and d will get fixed (since doing animation from worker > threads is pretty much the point). (c) yes and no, the animation part no, because it's a timing issue and switching processes messes with graphics's engines view of things. (d) no UIWebView is non-concurrent object, in future OS perhaps touching it from a background thread leads to a fatal exception (on purpose). > b - well of course you can't do UI work from FFI - you're not on the > UI thread - unless you call performSelectorOnMainThread:... from > there. no developers want to do that, also it's hard to determine what that is, using quicktime just to open a quicktime file not even display it? is that UI. Well it wasn't in the past, is now. > -- = = = ======================================================================== John M. McIntosh <[hidden email]> Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com = = = ======================================================================== _______________________________________________ Pharo-project mailing list [hidden email] http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project |
«
Return to Pharo Smalltalk Developers
|
1 view|%1 views
| Free forum by Nabble | Edit this page |