April 2017 - Pharo-users - lists.pharo.org

Re: [Pharo-users] Recommended stable Pharo 5.0 VM for Linux users?

by Esteban Lorenzano

> On 12 Apr 2017, at 11:43, Luke Gorrie <luke(a)snabb.co> wrote: > > On 12 April 2017 at 10:27, Esteban Lorenzano <estebanlm(a)gmail.com <mailto:estebanlm@gmail.com>> wrote: > yes, at the end I did a different process. > after cloning vm sources, now you do not need to generate themâ¦ just go to > > $ cd build.linuxWHATEVER/pharo.cog.spur/build > $ ./mvm > > that will work and produce the official latest VM :) > > Thanks for the information. I will give it a try. > > Any chance of updating the instructions in the README for future reference? It is in my TODO list. â¦ along with other milliards of things with more priority ;) I will welcome any Pull Request, in any area but specially on that one :D Esteban > >

April 12, 2017

Re: [Pharo-users] Traits, method flatten

by Hilaire

Investigating a bit on the mater, here is the conduct: - In a fresh 5.0, find an a version of my package where the Traits logic is not compromised (All traits methods are not duplicated in the classes using the traits) - Load this package, all seems fine. - Promote the class/trait category to a package, let's say TraitsPackage - Save this package locally in a .mcz file - In the Pharo5.0 image unload all my packages - Load back the TraitsPackage, all traits methods are duplicated... - Examine the source in the .mcz file: the traits logic is already compromised Is there anything wrong in this process? Conclusion1: Pharo5 does not know how to save Trait correctly. Conclusion2: Already get there, http://forum.world.st/Strange-behaviour-with-traits-td4891623.html Hilaire -- Dr. Geo http://drgeo.eu

April 12, 2017

Re: [Pharo-users] why is adding instance variables so slow?

by tesonep＠gmail.com

Hi, in non spur, the only improve that I think it can be made safely is moving the garbage collect operation to the migration of instances side, only executing the garbage collect if there are instances. In spur the garbage collect it is not needed anymore, because the way the become is working is not the same. In the old implementation, the old instances where in the image until they are removed by the garbage collector. So you can access them with the allInstances. In the new implementation, the old instances are marked as a forwarded to the new instances. So, there is no need for the garbage collect. This should be tried but I think it can work. Cheers, Pablo On Wed, Apr 12, 2017 at 1:17 PM, Guillermo Polito <guillermopolito(a)gmail.com > wrote: > > > On Wed, Apr 12, 2017 at 11:35 AM, Denis Kudriashov <dionisiydk(a)gmail.com> > wrote: > >> >> 2017-04-12 10:55 GMT+02:00 Guillermo Polito <guillermopolito(a)gmail.com>: >> >>> PharoClassInstaller>>migrateClasses: old to: new using: >>>> anInstanceModification >>>> instanceModification := anInstanceModification. >>>> old ifEmpty: [ ^ self ]. >>>> [ >>>> 1 to: old size do: [ :index | >>>> self updateClass: (old at: index) to: (new at: index)]. >>>> old elementsForwardIdentityTo: new. >>>> " Garbage collect away the zombie instances left behind in garbage >>>> memory in #updateInstancesFrom: " >>>> " If we don't clean up this garbage, a second update would revive them >>>> with a wrong layout! " >>>> " (newClass rather than oldClass, since they are now both newClass) " >>>> Smalltalk garbageCollect. >>>> ] valueUnpreemptively >>>> >>>> Commenting garbage collection here increases performance 10 times. >>>> Then commenting class update loop increases performance 3 times more. >>>> But this loop is required. It adopts all instances of old class to new one. >>>> And time here spent in #allInstances method. >>>> >>>> Can we remove manual garbage collection here? Why it is needed? >>>> >>> >>> Well, there is the comment that explains it and makes pretty good sense. >>> >> >> But is does not explain why these bad zombies exist. We investigates >> possible reasons and could not reproduce them. We will try remove garbage >> collection here in Pharo 7 >> > > No, this will break stuff! I'll try to explain what does it mean by zombie > instances to make some sense: > > - Imagine that you have class A + 10 instances of A. > > - We add an instance variable to A. > - this means the class builder will generate class A' that is the new > version of A. > - then, it migrates all instances of A to class A'. > This migration is not magic: > - 10 new instances of A' are created > - the state is migrated from the instances of A to A' > - each instance of A is becomed into its corresponding instance of > A' > - finally we become class A into A' > This step will make that old instances of A now have: > - the old format > - but point to the new class A > > If we do not garbage collect, this means that doing > > A allInstances > > will return not only the new 10 instances of A, but the old instances of > A'. > And that will break LOOOTS of stuff. > -- Pablo Tesone. tesonep(a)gmail.com

April 12, 2017

Re: [Pharo-users] why is adding instance variables so slow?

by Denis Kudriashov

2017-04-12 13:17 GMT+02:00 Guillermo Polito <guillermopolito(a)gmail.com>: > 1) each instance of A is becomed into its corresponding instance of A' > 2) finally we become class A into A' > This step will make that old instances of A now have: > - the old format > - but point to the new class A > step 1) ensures that there are no instances of class A anymore. Check following script: c1 := Class1 new. c2 := Class2 new. c1 becomeForward: c2. Class1 allInstances "=> #()". And full migration is executed in high priority uninterrupted process to ensure that between 1) and 2) nobody will instantiate Class1

April 12, 2017

Re: [Pharo-users] why is adding instance variables so slow?

by Guillermo Polito

On Wed, Apr 12, 2017 at 11:35 AM, Denis Kudriashov <dionisiydk(a)gmail.com> wrote: > > 2017-04-12 10:55 GMT+02:00 Guillermo Polito <guillermopolito(a)gmail.com>: > >> PharoClassInstaller>>migrateClasses: old to: new using: >>> anInstanceModification >>> instanceModification := anInstanceModification. >>> old ifEmpty: [ ^ self ]. >>> [ >>> 1 to: old size do: [ :index | >>> self updateClass: (old at: index) to: (new at: index)]. >>> old elementsForwardIdentityTo: new. >>> " Garbage collect away the zombie instances left behind in garbage >>> memory in #updateInstancesFrom: " >>> " If we don't clean up this garbage, a second update would revive them >>> with a wrong layout! " >>> " (newClass rather than oldClass, since they are now both newClass) " >>> Smalltalk garbageCollect. >>> ] valueUnpreemptively >>> >>> Commenting garbage collection here increases performance 10 times. >>> Then commenting class update loop increases performance 3 times more. >>> But this loop is required. It adopts all instances of old class to new one. >>> And time here spent in #allInstances method. >>> >>> Can we remove manual garbage collection here? Why it is needed? >>> >> >> Well, there is the comment that explains it and makes pretty good sense. >> > > But is does not explain why these bad zombies exist. We investigates > possible reasons and could not reproduce them. We will try remove garbage > collection here in Pharo 7 > No, this will break stuff! I'll try to explain what does it mean by zombie instances to make some sense: - Imagine that you have class A + 10 instances of A. - We add an instance variable to A. - this means the class builder will generate class A' that is the new version of A. - then, it migrates all instances of A to class A'. This migration is not magic: - 10 new instances of A' are created - the state is migrated from the instances of A to A' - each instance of A is becomed into its corresponding instance of A' - finally we become class A into A' This step will make that old instances of A now have: - the old format - but point to the new class A If we do not garbage collect, this means that doing A allInstances will return not only the new 10 instances of A, but the old instances of A'. And that will break LOOOTS of stuff.

April 12, 2017

Re: [Pharo-users] [ann] steering agile architecture video published by o'reilly

by Denis Kudriashov

Do you plan presentation for PharoDays? 2017-04-12 1:04 GMT+02:00 Tudor Girba <tudor(a)tudorgirba.com>: > Hi, > > I am happy to announce that OâReilly has published my Steering Agile > Architecture video training: > http://www.humane-assessment.com/blog/steering-agile- > architecture-video-training-oreilly/ > > This video is based on my work on humane assessment, and it features Moose > prominently. It also shows Pharo, GT and Roassal (in the last part). > > The video is indeed paid and it is mainly targeted to people that have a > Safari Books Online subscription. But, the interesting thing about it is > that it opens Pharo and Moose to a new kind of an audience, OâReilly being > the most prominent publishing company in the software area. > > Agile architecture is considered a prominent topic but the literature > (until now) was rather loose at how to make it really work in practice. > This course offers both a concrete approach based on humane assessment. As > you might know, humane assessment centers around the idea of building tools > cheaply as a way to figure the system out, and this is right now pretty > much only practicable in Moose and Pharo. > > One thing that I ask people when I go to conferences, the last one being > the Software Architecture Conference organized by OâReilly, is if they like > working with legacy systems. They actively dislike it. Then I show them > demos of how it can be like and then I ask them if they think these demos > are cool. And I never encountered one that did not find it cool. So, in a > way, we can now be quite confident in saying that we work on the single > platform that makes working with an existing system cool. Not a small feat. > > So, if you do get the chance to look at these things, you might also get a > concrete inspiration of how to leverage Pharo skills within companies that > do not yet work with Pharo. > > Cheers, > Doru > > > -- > www.tudorgirba.com > www.feenk.com > > "Presenting is storytelling." > > >

April 12, 2017

Re: [Pharo-users] Recommended stable Pharo 5.0 VM for Linux users?

by Luke Gorrie

On 12 April 2017 at 10:27, Esteban Lorenzano <estebanlm(a)gmail.com> wrote: > yes, at the end I did a different process. > after cloning vm sources, now you do not need to generate themâ¦ just go to > > $ cd build.linuxWHATEVER/pharo.cog.spur/build > $ ./mvm > > that will work and produce the official latest VM :) > Thanks for the information. I will give it a try. Any chance of updating the instructions in the README for future reference?

April 12, 2017

Re: [Pharo-users] why is adding instance variables so slow?

by Denis Kudriashov

2017-04-12 10:55 GMT+02:00 Guillermo Polito <guillermopolito(a)gmail.com>: > PharoClassInstaller>>migrateClasses: old to: new using: >> anInstanceModification >> instanceModification := anInstanceModification. >> old ifEmpty: [ ^ self ]. >> [ >> 1 to: old size do: [ :index | >> self updateClass: (old at: index) to: (new at: index)]. >> old elementsForwardIdentityTo: new. >> " Garbage collect away the zombie instances left behind in garbage memory >> in #updateInstancesFrom: " >> " If we don't clean up this garbage, a second update would revive them >> with a wrong layout! " >> " (newClass rather than oldClass, since they are now both newClass) " >> Smalltalk garbageCollect. >> ] valueUnpreemptively >> >> Commenting garbage collection here increases performance 10 times. >> Then commenting class update loop increases performance 3 times more. But >> this loop is required. It adopts all instances of old class to new one. And >> time here spent in #allInstances method. >> >> Can we remove manual garbage collection here? Why it is needed? >> > > Well, there is the comment that explains it and makes pretty good sense. > But is does not explain why these bad zombies exist. We investigates possible reasons and could not reproduce them. We will try remove garbage collection here in Pharo 7

April 12, 2017

Re: [Pharo-users] why is adding instance variables so slow?

by Guillermo Polito

On Wed, Apr 12, 2017 at 10:51 AM, Denis Kudriashov <dionisiydk(a)gmail.com> wrote: > Hi. > > I profiled a bit and found problem: > > PharoClassInstaller>>migrateClasses: old to: new using: > anInstanceModification > instanceModification := anInstanceModification. > old ifEmpty: [ ^ self ]. > [ > 1 to: old size do: [ :index | > self updateClass: (old at: index) to: (new at: index)]. > old elementsForwardIdentityTo: new. > " Garbage collect away the zombie instances left behind in garbage memory > in #updateInstancesFrom: " > " If we don't clean up this garbage, a second update would revive them > with a wrong layout! " > " (newClass rather than oldClass, since they are now both newClass) " > Smalltalk garbageCollect. > ] valueUnpreemptively > > Commenting garbage collection here increases performance 10 times. > Then commenting class update loop increases performance 3 times more. But > this loop is required. It adopts all instances of old class to new one. And > time here spent in #allInstances method. > > Can we remove manual garbage collection here? Why it is needed? > Well, there is the comment that explains it and makes pretty good sense. > > 2017-04-12 9:34 GMT+02:00 Peter Uhnak <i.uhnak(a)gmail.com>: > >> Hi, >> >> does anyone know why adding instance variables is so slow? >> >> I did some quick benchmarks (see below), resulting in more than order of >> magnitude speed difference between having it in the class definition and >> adding it later. >> >> In fact it is still much faster to delete the class and then recreate it >> with the instance variables in the d efinition. For four arguments it is >> till 8x faster to delete the class four times and recreate it then just add >> the variable. Unfortunately I cannot just trash the classes (they have >> methods and inheritance). >> >> So the question is: why is it so slow? can I somehow improve the >> performance? >> >> Thanks, >> Peter >> >> >> Benchmarks: >> >> >> [ >> cls := Object subclass: #Some1. >> cls removeFromSystem. >> ] bench. "'91 per second'" >> >> [ >> cls := Object subclass: #Some2 instanceVariableNames: 'variable'. >> cls removeFromSystem >> ] bench. "'90 per second'" >> >> [ >> cls := Object subclass: #Some3. >> cls addInstVarNamed: 'variable'. >> cls removeFromSystem. >> ] bench. "'7 per second'" >> >> [ >> cls := Object subclass: #Some4. >> cls removeFromSystem. >> cls := Object subclass: #Some4 instanceVariableNames: 'variable'. >> cls removeFromSystem. >> ] bench. "'43 per second'" >> >> >> >> >> [ >> cls := Object subclass: #Some3. >> cls addInstVarNamed: 'variable1'. >> cls addInstVarNamed: 'variable2'. >> cls addInstVarNamed: 'variable3'. >> cls addInstVarNamed: 'variable4'. >> cls removeFromSystem. >> ] bench. "'2 per second'" >> >> [ >> cls := Object subclass: #Some4. >> cls removeFromSystem. >> cls := Object subclass: #Some4 instanceVariableNames: 'variable1 >> variable2 variable3 variable4'. >> cls removeFromSystem. >> ] bench. "'44 per second'" >> >> [ >> cls := Object subclass: #Some5. >> cls removeFromSystem. >> cls := Object subclass: #Some5 instanceVariableNames: 'variable1'. >> cls removeFromSystem. >> cls := Object subclass: #Some5 instanceVariableNames: 'variable1 >> variable2'. >> cls removeFromSystem. >> cls := Object subclass: #Some5 instanceVariableNames: 'variable1 >> variable2 variable3'. >> cls removeFromSystem. >> cls := Object subclass: #Some5 instanceVariableNames: 'variable1 >> variable2 variable3 variable4'. >> cls removeFromSystem. >> ] bench. "'17.269 per second'" >> >> >

April 12, 2017

Re: [Pharo-users] why is adding instance variables so slow?

by Denis Kudriashov

Hi. I profiled a bit and found problem: PharoClassInstaller>>migrateClasses: old to: new using: anInstanceModification instanceModification := anInstanceModification. old ifEmpty: [ ^ self ]. [ 1 to: old size do: [ :index | self updateClass: (old at: index) to: (new at: index)]. old elementsForwardIdentityTo: new. " Garbage collect away the zombie instances left behind in garbage memory in #updateInstancesFrom: " " If we don't clean up this garbage, a second update would revive them with a wrong layout! " " (newClass rather than oldClass, since they are now both newClass) " Smalltalk garbageCollect. ] valueUnpreemptively Commenting garbage collection here increases performance 10 times. Then commenting class update loop increases performance 3 times more. But this loop is required. It adopts all instances of old class to new one. And time here spent in #allInstances method. Can we remove manual garbage collection here? Why it is needed? 2017-04-12 9:34 GMT+02:00 Peter Uhnak <i.uhnak(a)gmail.com>: > Hi, > > does anyone know why adding instance variables is so slow? > > I did some quick benchmarks (see below), resulting in more than order of > magnitude speed difference between having it in the class definition and > adding it later. > > In fact it is still much faster to delete the class and then recreate it > with the instance variables in the d efinition. For four arguments it is > till 8x faster to delete the class four times and recreate it then just add > the variable. Unfortunately I cannot just trash the classes (they have > methods and inheritance). > > So the question is: why is it so slow? can I somehow improve the > performance? > > Thanks, > Peter > > > Benchmarks: > > > [ > cls := Object subclass: #Some1. > cls removeFromSystem. > ] bench. "'91 per second'" > > [ > cls := Object subclass: #Some2 instanceVariableNames: 'variable'. > cls removeFromSystem > ] bench. "'90 per second'" > > [ > cls := Object subclass: #Some3. > cls addInstVarNamed: 'variable'. > cls removeFromSystem. > ] bench. "'7 per second'" > > [ > cls := Object subclass: #Some4. > cls removeFromSystem. > cls := Object subclass: #Some4 instanceVariableNames: 'variable'. > cls removeFromSystem. > ] bench. "'43 per second'" > > > > > [ > cls := Object subclass: #Some3. > cls addInstVarNamed: 'variable1'. > cls addInstVarNamed: 'variable2'. > cls addInstVarNamed: 'variable3'. > cls addInstVarNamed: 'variable4'. > cls removeFromSystem. > ] bench. "'2 per second'" > > [ > cls := Object subclass: #Some4. > cls removeFromSystem. > cls := Object subclass: #Some4 instanceVariableNames: 'variable1 > variable2 variable3 variable4'. > cls removeFromSystem. > ] bench. "'44 per second'" > > [ > cls := Object subclass: #Some5. > cls removeFromSystem. > cls := Object subclass: #Some5 instanceVariableNames: 'variable1'. > cls removeFromSystem. > cls := Object subclass: #Some5 instanceVariableNames: 'variable1 > variable2'. > cls removeFromSystem. > cls := Object subclass: #Some5 instanceVariableNames: 'variable1 > variable2 variable3'. > cls removeFromSystem. > cls := Object subclass: #Some5 instanceVariableNames: 'variable1 > variable2 variable3 variable4'. > cls removeFromSystem. > ] bench. "'17.269 per second'" > >

April 12, 2017