May 2018 - Pharo-dev - lists.pharo.org

New taskbar design proposal

by Cyril Ferlicot D.

Hi, I always had some problems with Pharo taskbar. The first one is that it is trasparent. It takes the whole width of Pharo but only the taskbar items are visible. This is bad because we have impression that we can click on the world at the bottom right of Pharo, but we cannot. Also I did not like the style. I think we can do something that looks more current. In order to ease modifications of the taskbar style I updated Pharo to add customization points to the theme for the taskbar. With the help of the community I also corrected some bugs in the pluggable button morph that made style changes hard. (The last one I found should be corrected in this PR: https://github.com/pharo-project/pharo/pull/1384) Now that it's easier to propose new designs I would like to propose a new one. It is present on this PR: https://github.com/pharo-project/pharo/pull/1443 You can see that change following this link. I added 2 GIF showing the result with the light and dark theme. I would like to have the opinion of the community before this PR is integrated/rejected. Have a nice day. -- Cyril Ferlicot https://ferlicot.fr

May 27, 2018

[Pharo 7.0-dev] Build #963: 21998 Cleanup Keymapping-Tests

by ci-pharo-ci-jenkins2＠inria.fr

There is a new Pharo build available! The status of the build #963 was: SUCCESS. The Pull Request #1442 was integrated: "21998 Cleanup Keymapping-Tests" Pull request url: https://github.com/pharo-project/pharo/pull/1442 Issue Url: https://pharo.fogbugz.com/f/cases/21998 Build Url: https://ci.inria.fr/pharo-ci-jenkins2/job/Test%20pending%20pull%20request%2…

May 27, 2018

[Pharo 7.0-dev] Build #962: 21961 move slot examples to a slot examples package

by ci-pharo-ci-jenkins2＠inria.fr

There is a new Pharo build available! The status of the build #962 was: SUCCESS. The Pull Request #1415 was integrated: "21961 move slot examples to a slot examples package " Pull request url: https://github.com/pharo-project/pharo/pull/1415 Issue Url: https://pharo.fogbugz.com/f/cases/21961 Build Url: https://ci.inria.fr/pharo-ci-jenkins2/job/Test%20pending%20pull%20request%2…

May 27, 2018

ZeroConf and Launcher

by Torsten Bergmann

ZeroConf (get.pharo.org) is nice - as it allows to easily setup a new machine with a fresh image. What about having the ability to download the latest Launcher using a zeroconf script too?

May 27, 2018

[Pharo 7.0-dev] Build #961: 21995-HeapTest-examples-should-use-String-streamContents-instead-of-Transcript

by ci-pharo-ci-jenkins2＠inria.fr

There is a new Pharo build available! The status of the build #961 was: FAILURE. The Pull Request #1438 was integrated: "21995-HeapTest-examples-should-use-String-streamContents-instead-of-Transcript" Pull request url: https://github.com/pharo-project/pharo/pull/1438 Issue Url: https://pharo.fogbugz.com/f/cases/21995 Build Url: https://ci.inria.fr/pharo-ci-jenkins2/job/Test%20pending%20pull%20request%2…

May 27, 2018

[Pharo 7.0-dev] Build #960: 21994-Generator-class-example-method-should-use-streamContents-instead-of-Transcript

by ci-pharo-ci-jenkins2＠inria.fr

There is a new Pharo build available! The status of the build #960 was: SUCCESS. The Pull Request #1437 was integrated: "21994-Generator-class-example-method-should-use-streamContents-instead-of-Transcript" Pull request url: https://github.com/pharo-project/pharo/pull/1437 Issue Url: https://pharo.fogbugz.com/f/cases/21994 Build Url: https://ci.inria.fr/pharo-ci-jenkins2/job/Test%20pending%20pull%20request%2…

May 27, 2018

[Pharo 7.0-dev] Build #959: 21993-ExternalBrowser-reference-to-Transcript-does-not-make-sense

by ci-pharo-ci-jenkins2＠inria.fr

There is a new Pharo build available! The status of the build #959 was: FAILURE. The Pull Request #1436 was integrated: "21993-ExternalBrowser-reference-to-Transcript-does-not-make-sense" Pull request url: https://github.com/pharo-project/pharo/pull/1436 Issue Url: https://pharo.fogbugz.com/f/cases/21993 Build Url: https://ci.inria.fr/pharo-ci-jenkins2/job/Test%20pending%20pull%20request%2…

May 27, 2018

Re: [Pharo-dev] Curious context change latency experiment results (wrt DelayScheduler)

by Ben Coman

On 6 May 2018 at 04:20, Eliot Miranda <eliot.miranda(a)gmail.com> wrote: > Hi Ben, > > > On May 5, 2018, at 7:25 AM, Ben Coman <btc(a)openinworld.com> wrote: > > > > On 5 May 2018 at 22:10, Ben Coman <btc(a)openinworld.com> wrote: > >> One of the key parts of Delay scheduling is setting the resumption time. >> There are two places this could be done. >> a. In Delay>>#schedule, running (for example) at userSchedulingPriority=40 >> b. In DelayMicrosecondScheduler>>#handleTImerEvent:, running at >> timingPriority=80 >> >> When we were using the millisecond clock for delay scheduling, >> it made sense to use (b.) since the clock might(?) roll over >> between when resumption time was set, and when #handleTImerEvent: >> expires delays. >> >> This should not be a problem now that we are using the microsecond clock, >> so I wondered what the latency penalty might be between (a.) and (b.). >> I ran a little experiment that indicates the cost/latency of switching >> threads, >> and was curious if anyone can comment on the validity of the experiment >> and interpretation of results. >> >> I made the following three changes... >> >> DelayScheduler subclass: #DelayMicrosecondScheduler >> instanceVariableNames: 'latencyStart countSinceLastZero' >> classVariableNames: 'LatencyCounts' >> poolDictionaries: '' >> category: 'Kernel-Processes' >> >> >> DelayMicrosecondScheduler>>#schedule: aDelay >> latencyStart:= Time primUTCMicrosecondsClock. "This is position (a.)" >> aDelay schedulerBeingWaitedOn ifTrue: [^self error: 'This Delay has >> already been scheduled.']. >> accessProtect critical: [ >> scheduledDelay := aDelay. >> timingSemaphore signal. "#handleTimerEvent: sets scheduledDelay:=nil" >> ]. >> >> >> DelayMicrosecondScheduler>>#handleTimerEvent: microsecondNowTick >> | microsecondNextTick | >> "Process any schedule requests" "This is position (b.)" >> scheduledDelay ifNotNil: [ >> |latency| >> latency := Time primUTCMicrosecondsClock - latencyStart. >> LatencyCounts ifNil: [ LatencyCounts := Bag new ]. >> LatencyCounts add: latency. >> latency = 0 >> ifTrue: [ countSinceLastZero := 1 + (countSinceLastZero ifNil: [0])] >> ifFalse: [Transcript >> crShow: 'zero latency count ' , countSinceLastZero printString ; >> show: ' before latency ', latency printString. >> countSinceLastZero := 0]. >> "Schedule the given delay." >> scheduledDelay scheduler: self resumptionTime: microsecondNowTick + (1000 >> * scheduledDelay millisecondDelayDuration). >> self scheduleDelay: scheduledDelay. >> scheduledDelay := nil ]. >> <snip> >> rest of method unchanged >> >> >> Then opened the Transcript and in Playground evaluated... >> Delay delaySchedulerClass: DelayMicrosecondScheduler. >> >> >> The Transcript results are shown below with some comments inserted. >> >> Now I guess the latency is affected by garbage collection. >> But one thing I was curious about is why the latency's were quantised in >> 1000s. >> >> Another interesting thing is that vast majority of the latency's were >> zero, >> which was a nice surprise, but can it be true? Or is it a consequence >> of the quantitisation rounding down? >> >> It seems that the idle-ness of the image affected how often a non-zero >> latency occurred. >> After I left the house for a while, the count of zero latency was very >> high, but a few still occurred. It would make sense there was less GC >> while idle. What is a good snippet of code to stress GC. I presume the >> latency might increase. >> >> >> zero latency count 2273 before latency 1000 >> zero latency count 943 before latency 1000 >> zero latency count 3666 before latency 1000 >> zero latency count 1643 before latency 1000 >> zero latency count 27 before latency 1000 >> "Left house for 20 minutes" >> zero latency count 12022 before latency 1000 >> zero latency count 15195 before latency 1000 >> zero latency count 41998 before latency 1000 >> "Returned from outing" >> zero latency count 128 before latency 1000 >> zero latency count 116 before latency 1000 >> zero latency count 555 before latency 1000 >> zero latency count 2377 before latency 1000 >> zero latency count 5423 before latency 1000 >> zero latency count 3178 before latency 1000 >> zero latency count 47 before latency 1000 >> zero latency count 2276 before latency 1000 >> "Left house to go shopping" >> zero latency count 6708 before latency 3000 >> zero latency count 4896 before latency 1000 >> zero latency count 433 before latency 1000 >> zero latency count 7106 before latency 1000 >> zero latency count 2195 before latency 1000 >> zero latency count 12397 before latency 1000 >> zero latency count 4815 before latency 2000 >> zero latency count 3480 before latency 1000 >> zero latency count 5690 before latency 1000 >> zero latency count 1670 before latency 1000 >> zero latency count 320 before latency 1000 >> zero latency count 1641 before latency 2000 >> zero latency count 290 before latency 1000 >> zero latency count 3835 before latency 1000 >> zero latency count 3818 before latency 1000 >> zero latency count 1132 before latency 1000 >> "Returned from shopping." >> zero latency count 2045 before latency 1000 >> zero latency count 671 before latency 1000 >> zero latency count 167 before latency 1000 >> zero latency count 147 before latency 1000 >> >> From the System Browser, inspecting the class variable LatencyCounts >> showed... >> 0 171329 >> 3000 1 >> 2000 3 >> 1000 49 >> > > P.S. after running a few more hours, LatencyCounts shows... > 0 1102691 > 1000 219 > 2000 12 > 3000 29 > 4000 21 > 5000 5 > 49000 1 > 75000 1 > > > It's likely that the vm code is using the cheap microsecond click which is > updated by the heartbeat, not the expensive one that interrogates the OS at > the instant. If not, it's the OS. I'll check soon. > I went looking and found an obvious source of quantitization multiplying by MicrosecondsPerMillisecond in the Win32 currentUTCMicroseconds() line 109 here... https://github.com/OpenSmalltalk/opensmalltalk-vm/blob/ e17db79411cfec767e04f3d94d12a642d920a30e/platforms/win32/vm/ sqWin32Time.c#L109 Running the same experiment (for 20 mins) on Linux produced a latency distribution in line with what I was originally expecting... usec count 1 135 2 992 3 3250 4 21161 5 17309 6 5341 7 1230 8 313 9 79 10 44 11 46 12 50 13 36 14 58 15 73 16 38 17 12 18 7 19 5 20 1 21 3 23 2 24 4 25 4 26 5 27 4 29 3 30 6 31 5 32 6 33 5 34 2 35 2 36 5 37 4 39 4 40 2 41 2 44 3 46 2 47 1 48 4 49 3 51 2 53 1 54 1 56 2 58 1 59 1 61 1 66 2 70 1 82 1 86 1 184 1 544 1 576 1 1212 1 1402 1 2016 1 self average. "4.8"microseconds self stdev. "12.9"microseconds self average + (5 * self stdev) "69.1"microseconds So there are nine outliers outside five sigma. @Clement, Would you expect your incremental GC to have much impact on those outliers? or its target pause is ~1000usecs anyway? cheers -ben

May 27, 2018

CI memory leak testing

by Ben Coman

On 26 May 2018 at 18:21, Tudor Girba <tudor(a)tudorgirba.com> wrote: > Hi, > > Yes, Glamour is stable since years. Yes, we did have some leak issues that > appeared along the way. Last year, we discovered that a memory leak came > from Rubric (specifically there was a global variable in it). People saw > inspectors laying around, so naturally, the issue was seen as being due to > GT. And now there is another one, which is still under investigation. For > me, the main issue is not the memory leak, but the fact that we have no > tool to reason about them. > I wonder if its possible/useful to define a generic mechanism that at the end of a CI test run all windows are force closed, garbage collected and then tallied objects the check a zero count for all classes except those explicitly defined, perhaps in a class manifest. May have picked this up earlier. cheers -ben

May 26, 2018

Re: [Pharo-dev] Crash on GC garbageCollect

by Ben Coman

On 26 May 2018 at 08:27, Stephan Eggermont <stephan(a)stack.nl> wrote: > Esteban Lorenzano <estebanlm(a)gmail.com> > wrote: > > > > > >> On 25 May 2018, at 23:48, Stephan Eggermont <stephan(a)stack.nl> wrote: > >> > >> Esteban Lorenzano <estebanlm(a)gmail.com> > >> wrote: > >>> thatâs because 6.1 images still comes with an old VM. > >>> I think I will backport the newer, but next week (testing time has > passed, I think). > >> > >> You probably want to include the TLS fix from this week for Win7 so > github > >> repos eork > > > > no I do not want :) > > because in that case I will need to repeat the full cycle: 2 weeks as > > latest, one month as stable in 7.0, than backport. > > and probably in that moment there will be another fix to include. > > I understand that you donât want, but releasing a stable vm for Pharo 6 > where Iceberg (that is, github) doesnât work on Win7 is also not really a > solution. > Waiting for the another stable vm is in the realm of "someday", chasing the length-of-a-piece-of-string. Better to make the improvement that can be made today, and follow up with the next asap after that. cheers -ben

May 26, 2018