Pharo-dev

By thread

pharo-dev@lists.pharo.org

3 participants
144616 messages

Re: [Pharo-dev] Beacon moved to github

by Pavel Krivanek

You probably cannot set it to complete read-only mode, but you can disallow in the project settings "Public write access" and then remove all contributors. -- Pavel 2017-07-07 16:45 GMT+02:00 Norbert Hartl <norbert(a)hartl.name>: > I used the git migration tool of Peter Uhnak to migrate Beacon from > smalltalkhub to github. It is available now here > > https://github.com/pharo-project/pharo-beacon > > I just need to know how I can switch a smalltalkhub project to read-only. > > FYI, > > Norbert > >

July 7, 2017

Re: [Pharo-dev] Beacon moved to github

by Denis Kudriashov

Hi. I though it was already done :) One notice. Should we all follow standard convention on repo structure? Because source directory in beacon is called "repository" instead "src" like in other repos. 2017-07-07 16:45 GMT+02:00 Norbert Hartl <norbert(a)hartl.name>: > I used the git migration tool of Peter Uhnak to migrate Beacon from > smalltalkhub to github. It is available now here > > https://github.com/pharo-project/pharo-beacon > > I just need to know how I can switch a smalltalkhub project to read-only. > > FYI, > > Norbert > >

July 7, 2017

Beacon moved to github

by Norbert Hartl

I used the git migration tool of Peter Uhnak to migrate Beacon from smalltalkhub to github. It is available now here https://github.com/pharo-project/pharo-beacon <https://github.com/pharo-project/pharo-beacon> I just need to know how I can switch a smalltalkhub project to read-only. FYI, Norbert

July 7, 2017

Re: [Pharo-dev] [ANN] Iceberg 0.5.1 with Pull Request review tool

by Esteban Lorenzano

Hi, just a little update: yesterday I realised the tool lacked an important view: the changes between the PR and the actual destination of it (for example, you are doing a PR into âdevelopmentâ branch, and what is there could be different to what is loaded in your image. So I added that tab in the changes browser for the Pull Request: âChanges vs. destinationâ : this should allow you to analyse the PR in a more consistent way as before. now you have: - changes vs. destination - changes since selection (show changes against image version) - changes vs. parent (show changes between selected commit and itâs parent) that should be enough to actually review the pull request ;) Esteban > On 7 Jul 2017, at 08:37, Esteban Lorenzano <estebanlm(a)gmail.com <mailto:estebanlm@gmail.com>> wrote: > > >> On 6 Jul 2017, at 17:35, Tim Mackinnon <tim(a)testit.works <mailto:tim@testit.works>> wrote: >> >> Will this update be picked up with the scripted steps you described for 0.5 (particularly if you want to run in a v6 image?).And do we just run them again to get it? (I assume that if you have already loaded 0.5 you just do the metacello bit - but in a new image, would those steps get you 0.5.1?) > > yes (because 0.5.1 is released so integrated on master). > otherwise you can specify version by doing: > > Metacello new > baseline: 'Iceberg'; > repository: 'github://pharo-vcs/iceberg <github://pharo-vcs/iceberg>:v0.5.1â; > load. > > Esteban > >> >> (e.g. your wrote: >> >> then, to update, execute this (sorry, this is like that because we have still an older Metacello version): >> >> #('Iceberg-UI' 'Iceberg-Plugin' 'Iceberg-Metacello-Integration' 'Iceberg-Libgit' 'Iceberg' 'BaselineOfIceberg' 'LibGit-Core' 'BaselineOfLibGitâ) >> do: [ :each | each asPackage removeFromSystem ]. >> >> Metacello new >> baseline: 'Iceberg'; >> repository: 'github://pharo-vcs/iceberg <github://pharo-vcs/iceberg>'; >> load. >> ) >> >> Tim >> >>> On 6 Jul 2017, at 16:23, Esteban Lorenzano <estebanlm(a)gmail.com <mailto:estebanlm@gmail.com>> wrote: >>> >>> it should be fixed now, I think :) >>> >>>> On 6 Jul 2017, at 17:07, Pavel Krivanek <pavel.krivanek(a)gmail.com <mailto:pavel.krivanek@gmail.com>> wrote: >>>> >>>> There seems to be an issue in the baseline with circular dependencies. Load it, skip all warnings and then try to reload all packages in the Iceberg repository (from the Iceberg window). >>>> >>>> -- Pavel >>>> >>>> 2017-07-06 16:53 GMT+02:00 Norbert Hartl <norbert(a)hartl.name <mailto:norbert@hartl.name>>: >>>> Am I the only one or are there some classes missing? >>>> >>>> Norbert >>>> >>>>> Am 06.07.2017 um 15:35 schrieb Esteban Lorenzano <estebanlm(a)gmail.com <mailto:estebanlm@gmail.com>>: >>>>> >>>>> Hi, >>>>> >>>>> I just released Iceberg version 0.5.1 with a Pull Request tool Guille and I worked on since yesterday. >>>>> It allows you to list open Pull Requests (by right click on a repo, GitHub/Review pull requestsâ¦ option): >>>>> >>>>> <Screen Shot 2017-07-06 at 14.59.32.png> >>>>> >>>>> And then if you doubleclick on one (or select it with right button), you will see this: >>>>> >>>>> >>>>> <Screen Shot 2017-07-06 at 15.00.56.png> >>>>> it allows you to see changes and >>>>> >>>>> - merge changes into your image (in case you want to see more in details the code, run tests, etc.) >>>>> - accept a pull request >>>>> - reject a pull request >>>>> >>>>> no, it does not shows (at least *yet*) comments, and it does not allows you to add comments, reviews, etc. >>>>> this could be done, but not time to implement it now, so for now this has to be enough. >>>>> >>>>> Again, this can be loaded in a 6.0 image by executing this script: >>>>> >>>>> #('Iceberg-UI' 'Iceberg-Plugin' 'Iceberg-Metacello-Integration' 'Iceberg-Libgit' 'Iceberg' 'BaselineOfIceberg' 'LibGit-Core' 'BaselineOfLibGit') do: [ :each | each asPackage removeFromSystem ]. >>>>> >>>>> Metacello new >>>>> baseline: 'Iceberg'; >>>>> repository: 'github://pharo-vcs/iceberg:v0.5.1 <>'; >>>>> load. >>>>> >>>>> (and you still need to have the vm that is meant for Pharo7) >>>>> >>>>> This tools are open for you to use on your projectsâ¦ and to improve them, I accept pull requests on pharo-vcs/iceberg. >>>>> >>>>> cheers! >>>>> Esteban >>>>> >>>>> >>>> >>>> >>> >> >

July 7, 2017

Re: [Pharo-dev] [ANN] Iceberg 0.5.1 with Pull Request review tool

by Esteban Lorenzano

> On 6 Jul 2017, at 17:35, Tim Mackinnon <tim(a)testit.works> wrote: > > Will this update be picked up with the scripted steps you described for 0.5 (particularly if you want to run in a v6 image?).And do we just run them again to get it? (I assume that if you have already loaded 0.5 you just do the metacello bit - but in a new image, would those steps get you 0.5.1?) yes (because 0.5.1 is released so integrated on master). otherwise you can specify version by doing: Metacello new baseline: 'Iceberg'; repository: 'github://pharo-vcs/iceberg <github://pharo-vcs/iceberg>:v0.5.1â; load. Esteban > > (e.g. your wrote: > > then, to update, execute this (sorry, this is like that because we have still an older Metacello version): > > #('Iceberg-UI' 'Iceberg-Plugin' 'Iceberg-Metacello-Integration' 'Iceberg-Libgit' 'Iceberg' 'BaselineOfIceberg' 'LibGit-Core' 'BaselineOfLibGitâ) > do: [ :each | each asPackage removeFromSystem ]. > > Metacello new > baseline: 'Iceberg'; > repository: 'github://pharo-vcs/iceberg <github://pharo-vcs/iceberg>'; > load. > ) > > Tim > >> On 6 Jul 2017, at 16:23, Esteban Lorenzano <estebanlm(a)gmail.com <mailto:estebanlm@gmail.com>> wrote: >> >> it should be fixed now, I think :) >> >>> On 6 Jul 2017, at 17:07, Pavel Krivanek <pavel.krivanek(a)gmail.com <mailto:pavel.krivanek@gmail.com>> wrote: >>> >>> There seems to be an issue in the baseline with circular dependencies. Load it, skip all warnings and then try to reload all packages in the Iceberg repository (from the Iceberg window). >>> >>> -- Pavel >>> >>> 2017-07-06 16:53 GMT+02:00 Norbert Hartl <norbert(a)hartl.name <mailto:norbert@hartl.name>>: >>> Am I the only one or are there some classes missing? >>> >>> Norbert >>> >>>> Am 06.07.2017 um 15:35 schrieb Esteban Lorenzano <estebanlm(a)gmail.com <mailto:estebanlm@gmail.com>>: >>>> >>>> Hi, >>>> >>>> I just released Iceberg version 0.5.1 with a Pull Request tool Guille and I worked on since yesterday. >>>> It allows you to list open Pull Requests (by right click on a repo, GitHub/Review pull requestsâ¦ option): >>>> >>>> <Screen Shot 2017-07-06 at 14.59.32.png> >>>> >>>> And then if you doubleclick on one (or select it with right button), you will see this: >>>> >>>> >>>> <Screen Shot 2017-07-06 at 15.00.56.png> >>>> it allows you to see changes and >>>> >>>> - merge changes into your image (in case you want to see more in details the code, run tests, etc.) >>>> - accept a pull request >>>> - reject a pull request >>>> >>>> no, it does not shows (at least *yet*) comments, and it does not allows you to add comments, reviews, etc. >>>> this could be done, but not time to implement it now, so for now this has to be enough. >>>> >>>> Again, this can be loaded in a 6.0 image by executing this script: >>>> >>>> #('Iceberg-UI' 'Iceberg-Plugin' 'Iceberg-Metacello-Integration' 'Iceberg-Libgit' 'Iceberg' 'BaselineOfIceberg' 'LibGit-Core' 'BaselineOfLibGit') do: [ :each | each asPackage removeFromSystem ]. >>>> >>>> Metacello new >>>> baseline: 'Iceberg'; >>>> repository: 'github://pharo-vcs/iceberg:v0.5.1 <>'; >>>> load. >>>> >>>> (and you still need to have the vm that is meant for Pharo7) >>>> >>>> This tools are open for you to use on your projectsâ¦ and to improve them, I accept pull requests on pharo-vcs/iceberg. >>>> >>>> cheers! >>>> Esteban >>>> >>>> >>> >>> >> >

July 7, 2017

Re: [Pharo-dev] Discussing the roadmap

by Stephane Ducasse

On Thu, Jul 6, 2017 at 11:27 PM, Tim Mackinnon <tim(a)testit.works> wrote: > I didnât mean to touch a nerve - and this was why I wrote âminorâ points - but you did ask for feedbackâ¦ you did not :) Just that if we just count on fully book engineers working on super important features we will not make it. This is why once in a while I sit and fix a boring pop up or related. I hate the stupid pop up for pakcgae selection and one day I will have to retry to kill it. > > Point noted on giving user feedback - Iâd actually like to fix things, but currently its just too hard to submit fixes other than pull requests for documentation that is sitting in git hub (and I have submitted those). I know you are working on a simpler way to contribute (and getting the GitHub process smoothed out is in progress so its not fair to comment on that yet as its early days - and much appreciated. > > The Alt-Tab issue in the tracker since 2015 - but I will try and report others and possibly Calypso might have some better keyboard navigation tricks that help. do you have the bug entry? > > For website documentation - Iâll take it back, as I went back again just now and everything I could think of was there (a lot in the news section which I hadnât noticed). I do think the download page is a bit confusing but I know there is work being done on that. > > Tim > > >> On 6 Jul 2017, at 19:11, Stephane Ducasse <stepharo.self(a)gmail.com> wrote: >> >>> Only 2 minor items stick out as missing: >>> >>> 1) Continuing to improve keyboard shortcut support (its a lot better, but >>> not quite completed - I really miss some shortcuts, particularly the ability >>> to meta-tab between windows - ALT-Tab only works in some windows, and widen >>> selection in the editor - e.g. CMD-W in IntelliJ to increasing select a >>> word, a statement, a function etc) >> >> There are three issues there: >> - moving from the vm the keyboard generation (this will be done via >> OSWindow and the VM can focus its main job: execution). >> - then there is the way we declare/manage shortcuts: this is an >> ongoing efforts where everybody can join. We should remove the >> hardcoded shortcut and turn them into KeyBindings >> - Most of the tools will be throw away when bloc will be integrated. >> Now it does not mean that we should not do it. >> >> Now an important point: as a user you should report but you can also >> contribute. >> This is super difficult for me to fix something that I have no clue >> how to reproduce/never encounter. >> You see the mental model of a real open-source movement is to share >> but also to produce together and the point is >> what is the reward model: why should I spend time on pain I do not >> feel. Of course we are already doing a lot >> of actions that is not focus on our immediate needs and we should >> improve the situation but I wanted to share >> with you this perspective. >> >> >>> >>> 2) Keeping the website documentation more up to date (again the pharo.org >>> website is very slick, and a great showcase, however often there are late >>> breaking changes which new users wonât know about unless they trace through >>> the news groups or subscribe to some blogs). If we could also focus on >>> keeping it simple but up to date - that might help. >> >> Where is it not up to date? >> We are always reacting to any broken piece there. >> You see producing Pharo by Example 5 was A LOT of work and I thank >> again Nicolai Hesse for all his huge efforts. >> >> >>> I also have 1 small query - the references to the text editor, is this >>> making the editor that was demoâd at PharoDays 2107 integrated? As that one >>> was very cool? >> >> Yes calypso will replace our old friend Nautilus and we will certainly >> need to put a lot of love into calypso to get >> it to the right level. Now the basis is sound. >> About the node we will revisit smart suggestions and indeed sometimes >> there are problems to catch the correct AST nodes. >> >>> I donât know how extensible it is (Iâm hoping so - and that >>> it might help making it easier to improve code editing tools such that >>> refactorings can more intelligently understand where your cursor is with >>> regards to the AST node underneath it, or we can add more intellisense >>> options that match some of the things we see in IntelliJ). >>> >>> Tim >>> >>> On 6 Jul 2017, at 13:00, Pavel Krivanek <pavel.krivanek(a)gmail.com> wrote: >>> >>> >>> TxText removal is already done too. >>> >>> -- Pavel >>> >>> 2017-07-06 10:55 GMT+02:00 Stephane Ducasse <stepharo.self(a)gmail.com>: >>>> >>>> We would like to share this list with you and get your feedback and >>>> inputs. >>>> It will be presented and discussed again at ESUG. >>>> >>>> Stef on the behalf of the engineering team of the Pharo consortium. >>>> >>>> # Pharo 7 (and 8) potential roadmap >>>> >>>> This document contains a list of actions that should be done during >>>> Pharo 7 and 8. >>>> It is not a definitive list. It is the result of a discussion within >>>> the engineer team and RModers. >>>> It first lists actions that should be performed at the image level >>>> then lists the actions for the VM. >>>> >>>> >>>> ## Image >>>> As a general principle, we will try to remove something when we add a >>>> new features or version of a component. >>>> >>>> ### New tools >>>> - Iceberg: Iceberg is the new tool suite to handle new VCS in Pharo. >>>> For the moment it supports Git. This tool will become central to the >>>> development of projects in Pharo. The key and first enhancements will >>>> be: >>>> - cherry picking >>>> - multiple directories support, subtrees >>>> - better new development process support >>>> >>>> - Calypso: Calypso is a new tool suite for editing and navigating >>>> code. It is modular and can easily be extended. >>>> - integration, set as the default browser >>>> >>>> - Hermes (binary loader): Hermes is a code loader that does not >>>> require the compiler to be loaded. It will be used to speed up the >>>> bootstrap. >>>> >>>> - Beacon: Beacon is an announcement-based logger. It should be >>>> introduced but in addition the left over of previous logging should be >>>> cleaned and removed for example many of the transcript show: should be >>>> removed. >>>> >>>> - Inspector refreshing: the inspector should refresh its values by >>>> default. >>>> >>>> - Cargo: Cargo is a new package manager. It supports the expression of >>>> dependencies between packages. We are currently validating that it can >>>> support conditional loading (platform) and building new tooling to >>>> express and validate data. >>>> >>>> - Check dependencies when committing. Pharo comes with a dependency >>>> analyser tools. Such tools should be used before commiting to warn the >>>> user when a new dependency is introduced in a package. >>>> >>>> ### Cleaning bloat >>>> - Removing of Nautilus: Once Calypso will be integrated and exhibit >>>> similar fetaures than Nautilus. Nautilus should be removed. >>>> - Remove old text editor: there is only one or two widgets still using >>>> the old text editor and the rubric text editor. >>>> - Remove TxText: TxText was an attempt to get a new generation text >>>> editor. Now it has been rewritten with a new design in Bloc so we >>>> should remove it since Bloc editor is better and actively maintained. >>>> - Remove Komitter: Iceberg already supports cherrypicking on commit >>>> therefore Komitter can be safely removed from Pharo. >>>> - Remove system categorizer: The old system categorizer is not used >>>> anymore and should be remove. >>>> >>>> - Old compiler removal: The old compiler should now get unloaded from >>>> Pharo. >>>> - The old compiler should be moved to an external package and make >>>> sure that it can be reloaded. >>>> - In addition the encoders should be separated. (@@ more here@@) >>>> >>>> - Old inspector cleanup: we should remove the eyeInspector framework >>>> but we should introduce a minimal fallback inspector. This inspector >>>> should work without Spec or any frameworks. >>>> >>>> - Clean behavior protocol. The number of methods in the core Behavior, >>>> ClassDescription and Class requires some analysis and cleaning. >>>> >>>> - Kernel modularization: the effort on modularising the packages >>>> should be continued. >>>> >>>> ### Language infrastructure >>>> The following points are more related to the infrastructure of >>>> manipulating and loading class definitions. There are the basis for >>>> the future module system and cleaning some often hidden parts of the >>>> system. >>>> >>>> - New class definition: The class definition is not scaling anymore >>>> due to the large number of options (traits, slot, tage, package, >>>> immediate, ephemeron). A fluid-based message API should be designed. >>>> >>>> - Support for Undefined classes: When loading old code or code whose >>>> superclass has not yet being loaded, the system has inconsistent >>>> behavior. Depending on the tools, it may not load the code, raise a >>>> warning or decide that the superclass is Object without any other >>>> notice and losing the name of the original superclass. We are working >>>> on a new mechanism to support a unique way to handle such case. To be >>>> presented at IWST/ESUG >>>> >>>> - Class definition parser: Class definition is parser in different >>>> place of the system and in addition the output is the direct creation >>>> of a class object instead of an object representing the class >>>> definition that can be further manipulated. We are working on class >>>> definition parser. It produces a separate AST. It will help the >>>> building of tools. >>>> >>>> - Better update infrastructure. Pablo Tesone has been working on a >>>> better update mechanism, better modular class builder. >>>> >>>> - Ring2: Ring is a metamodel to manipulate code that is not actively >>>> running in the current system. It is useful to perform analysis (such >>>> as browsing, navigating off-line or remote definitions) before >>>> actually loading the code. Ring got redesigned by Pavel Krivanek to >>>> support the bootstrap of Pharo 60. The results is Ring2. >>>> >>>> - Opal is the Pharo Compiler. It should be enhanced to support handle >>>> better the warning and a general enhancement pass is needed. >>>> >>>> ### System enhancements >>>> >>>> - Commandline enhancements. RMOD is currently improving the >>>> command-line infrastructure and making sure that the system can work >>>> in read-only mode. >>>> >>>> - Cleaning streams: the idea is to make sure that the system does not >>>> use the old streams. The idea is to start using the fileStreams and >>>> make sure that the Stream package can be substituated in the future by >>>> other streams with the same API. Therefore hardcoded class names >>>> should be substituted with factory (readSream writeStream). From that >>>> perspective we do not think that it is wise to introduce Xtreams. We >>>> should analyse the API of the current implementation. >>>> >>>> - New theme from the beginning. It is really important that each >>>> version of Pharo is identified open the default opening. Pharo should >>>> come up with two default themes: one light and one dark. >>>> >>>> - Better themes/palettes support >>>> - Better and nicer Tabs. Tabs design should be revisited. >>>> >>>> - OSWindow world rendering: the effort to remove the logic of the >>>> windowing from the VM should be continued. OSWindow should be used >>>> instead. >>>> >>>> - Freetype: The current freetype plugin is the source of many bugs and >>>> problem. Thibault Raffaillac used FFI to do direct call to openGL >>>> (@@to verify@@). >>>> >>>> >>>> - Refactoring 2: Refactoring 2 is an improved version of the >>>> refactorings developed by Gustavos Santos and they should used to >>>> replace the existing one. >>>> >>>> >>>> ## VM >>>> >>>> - 64-bits by default: Mac and Linux 64 bits VMs have been working for >>>> over a year and since June 2017 a Windows 64 bits VM has been working. >>>> The plan is to stabilize each part of Pharo that is not working in 64 >>>> bits as well as in 32 bits and make the 64 bits Pharo images and VM >>>> the default download for Pharo. >>>> >>>> - Headless VM: Ronie Salgado has built a headless VM on his VM branch, >>>> we need to merge and check everything works fine. With the headless >>>> VM, Pharo can be run in true headless mode (not with a hidden window) >>>> and it is required for future VM features (embeddable VM, ...). >>>> >>>> - Embeddable VM: The VM should be able to be embedded in external >>>> applications, the application accessing objects through well-defined >>>> APIs. >>>> >>>> - Idle VM: The goal is to avoid the active waiting loop consuming >>>> several per cent of cpu time when the Pharo image is not doing >>>> anything. >>>> >>>> - Android VM: integration of the Android VM build and tests in the >>>> integration setup. >>>> >>>> - Threaded FFI: all FFI calls should be non-blocking (reviving >>>> prototype of Eliot) >>>> >>>> - ZeroConf for ARM: The ARM VM should be available from the zeroConf >>>> servers >>>> >>>> - Better support for large heaps (GC tuning API, incremental GC). >>>> Pharo 64 bit is now able to manage large heap. However better >>>> performance can be proposed by offering better settings for the >>>> different GC zone. >>>> >>>> - Integrate various fixes to support better high resolution display. >>>> In Retina mac display Pharo looks blurry. The fix to support Retina >>>> should be integrated in the VM. >>>> >>>> ### Sista-related >>>> Sista is an optimizing JIT for Pharo. It is the result of multiple >>>> years of developement by Clement Bera and Eliot Miranda. An optimizing >>>> JIT is manipulating code (folding constant, rewriting it) before >>>> compiling it to assembly code. >>>> >>>> - New Block Closure implementation by default: Allows one to implement >>>> in the Opal compiler the copying and clean blocks optimisations, >>>> reduce massively the complexity of some parts of the JIT (debugging >>>> API to map machine code pc to bytecode pc for example) and is required >>>> for the Sista support. Some work remains in debugging/IDE support. >>>> >>>> - New Bytecode set in production: Eases bytecode decoding (simplifying >>>> the code base of the bytecode to source code decompiler, the debugger, >>>> the JIT, etc.), lifts compiled code limitations (size of jumps, etc.) >>>> and is required for the Sista support. Some work remains in debugging >>>> support. >>>> >>>> - Read-only objects: They work in Pharo 6, but the in-image support >>>> needs to be improved (fall-back code of primitives, etc.) and some >>>> polishing needs to be done (primitive error code not always correct, >>>> etc.). Used for the support of different project, including GBS. >>>> >>>> - Literals immutability (based on read-only objects). In particular an >>>> analysis of the image usage of literal (string mutation) is needed. >>>> >>>> - Sista preview: A first version of Sista will be integrated, yielding >>>> 1.5x performance boost on most applications, but will be optional and >>>> will require specific constraints (not toying too much with literal >>>> mutability, partial IDE support, etc.). >>>> >>>> >>>> ### Already done >>>> >>>> - [DONE] Remove Shoreline reporter. >>>> - [DONE] Kernel should not depend on Traits: This will speed up the >>>> bootstrap and support the modular introduction of alternate traits >>>> implementation currently designed by Pablo Tesone and tested in the >>>> new generation of Moose. >>>> >>> >>> >> > >

July 7, 2017

Re: [Pharo-dev] Discussing the roadmap

by K K Subbu

On Friday 07 July 2017 03:26 AM, Eliot Miranda wrote: > Let's say that currently the mark-compact collector collects at about > 1Gb per second (that's about what I'm seeing on my MacMini for a 600Mb > heap, collected and compacted in about 530ms on a 2.3GHz Core i7, = > 1.1Gb/s). For the moment let's assume the GC scales linearly. Then a > 10Gb server application will occasionally suffer 10 second pauses, which > is unacceptable for any kind of high-availability server. The total > execution time overhead of GC may be in the 1%-10% range, but the pause > times may be unacceptable (note that they're /not/ unacceptable for my > employer's batch usage). Good point! Object memory is a shared resource and like any shared resource will eventually become a bottleneck as it grows. We could also tackle the problem of large apps with a different approach. The overheads of VM process may become negligible enough to make message passing across multiple co-operating processes more efficient than within a large monolithic process. E.g. if a computation of 1GB of data generates 10GB of garbage, then the extra 10GB need not be explicitly garbage collected if the task is done in a child (worker) process and terminated after the computed result is passed back to its parent. Regards .. Subbu

July 7, 2017

Re: [Pharo-dev] Discussing the roadmap

by Sven Van Caekenberghe

> On 6 Jul 2017, at 23:56, Eliot Miranda <eliot.miranda(a)gmail.com> wrote: > > Hi Sven, > > On Thu, Jul 6, 2017 at 10:10 AM, Sven Van Caekenberghe <sven(a)stfx.eu> wrote: > > > On 6 Jul 2017, at 18:41, Eliot Miranda <eliot.miranda(a)gmail.com> wrote: > > > > Hi Stef, > > > >> On Jul 6, 2017, at 1:55 AM, Stephane Ducasse <stepharo.self(a)gmail.com> wrote: > >> > >> We would like to share this list with you and get your feedback and inputs. > >> It will be presented and discussed again at ESUG. > >> > >> Stef on the behalf of the engineering team of the Pharo consortium. > >> > >> # Pharo 7 (and 8) potential roadmap > >> > >> This document contains a list of actions that should be done during > >> Pharo 7 and 8. > >> It is not a definitive list. It is the result of a discussion within > >> the engineer team and RModers. > >> It first lists actions that should be performed at the image level > >> then lists the actions for the VM. > >> > >> > >> ## Image > >> As a general principle, we will try to remove something when we add a > >> new features or version of a component. > >> > >> ### New tools > >> - Iceberg: Iceberg is the new tool suite to handle new VCS in Pharo. > >> For the moment it supports Git. This tool will become central to the > >> development of projects in Pharo. The key and first enhancements will > >> be: > >> - cherry picking > >> - multiple directories support, subtrees > >> - better new development process support > >> > >> - Calypso: Calypso is a new tool suite for editing and navigating > >> code. It is modular and can easily be extended. > >> - integration, set as the default browser > >> > >> - Hermes (binary loader): Hermes is a code loader that does not > >> require the compiler to be loaded. It will be used to speed up the > >> bootstrap. > >> > >> - Beacon: Beacon is an announcement-based logger. It should be > >> introduced but in addition the left over of previous logging should be > >> cleaned and removed for example many of the transcript show: should be > >> removed. > >> > >> - Inspector refreshing: the inspector should refresh its values by default. > >> > >> - Cargo: Cargo is a new package manager. It supports the expression of > >> dependencies between packages. We are currently validating that it can > >> support conditional loading (platform) and building new tooling to > >> express and validate data. > >> > >> - Check dependencies when committing. Pharo comes with a dependency > >> analyser tools. Such tools should be used before commiting to warn the > >> user when a new dependency is introduced in a package. > >> > >> ### Cleaning bloat > >> - Removing of Nautilus: Once Calypso will be integrated and exhibit > >> similar fetaures than Nautilus. Nautilus should be removed. > >> - Remove old text editor: there is only one or two widgets still using > >> the old text editor and the rubric text editor. > >> - Remove TxText: TxText was an attempt to get a new generation text > >> editor. Now it has been rewritten with a new design in Bloc so we > >> should remove it since Bloc editor is better and actively maintained. > >> - Remove Komitter: Iceberg already supports cherrypicking on commit > >> therefore Komitter can be safely removed from Pharo. > >> - Remove system categorizer: The old system categorizer is not used > >> anymore and should be remove. > >> > >> - Old compiler removal: The old compiler should now get unloaded from Pharo. > >> - The old compiler should be moved to an external package and make > >> sure that it can be reloaded. > >> - In addition the encoders should be separated. (@@ more here@@) > >> > >> - Old inspector cleanup: we should remove the eyeInspector framework > >> but we should introduce a minimal fallback inspector. This inspector > >> should work without Spec or any frameworks. > >> > >> - Clean behavior protocol. The number of methods in the core Behavior, > >> ClassDescription and Class requires some analysis and cleaning. > >> > >> - Kernel modularization: the effort on modularising the packages > >> should be continued. > >> > >> ### Language infrastructure > >> The following points are more related to the infrastructure of > >> manipulating and loading class definitions. There are the basis for > >> the future module system and cleaning some often hidden parts of the > >> system. > >> > >> - New class definition: The class definition is not scaling anymore > >> due to the large number of options (traits, slot, tage, package, > >> immediate, ephemeron). A fluid-based message API should be designed. > >> > >> - Support for Undefined classes: When loading old code or code whose > >> superclass has not yet being loaded, the system has inconsistent > >> behavior. Depending on the tools, it may not load the code, raise a > >> warning or decide that the superclass is Object without any other > >> notice and losing the name of the original superclass. We are working > >> on a new mechanism to support a unique way to handle such case. To be > >> presented at IWST/ESUG > >> > >> - Class definition parser: Class definition is parser in different > >> place of the system and in addition the output is the direct creation > >> of a class object instead of an object representing the class > >> definition that can be further manipulated. We are working on class > >> definition parser. It produces a separate AST. It will help the > >> building of tools. > >> > >> - Better update infrastructure. Pablo Tesone has been working on a > >> better update mechanism, better modular class builder. > >> > >> - Ring2: Ring is a metamodel to manipulate code that is not actively > >> running in the current system. It is useful to perform analysis (such > >> as browsing, navigating off-line or remote definitions) before > >> actually loading the code. Ring got redesigned by Pavel Krivanek to > >> support the bootstrap of Pharo 60. The results is Ring2. > >> > >> - Opal is the Pharo Compiler. It should be enhanced to support handle > >> better the warning and a general enhancement pass is needed. > >> > >> ### System enhancements > >> > >> - Commandline enhancements. RMOD is currently improving the > >> command-line infrastructure and making sure that the system can work > >> in read-only mode. > >> > >> - Cleaning streams: the idea is to make sure that the system does not > >> use the old streams. The idea is to start using the fileStreams and > >> make sure that the Stream package can be substituated in the future by > >> other streams with the same API. Therefore hardcoded class names > >> should be substituted with factory (readSream writeStream). From that > >> perspective we do not think that it is wise to introduce Xtreams. We > >> should analyse the API of the current implementation. > >> > >> - New theme from the beginning. It is really important that each > >> version of Pharo is identified open the default opening. Pharo should > >> come up with two default themes: one light and one dark. > >> > >> - Better themes/palettes support > >> - Better and nicer Tabs. Tabs design should be revisited. > >> > >> - OSWindow world rendering: the effort to remove the logic of the > >> windowing from the VM should be continued. OSWindow should be used > >> instead. > >> > >> - Freetype: The current freetype plugin is the source of many bugs and > >> problem. Thibault Raffaillac used FFI to do direct call to openGL > >> (@@to verify@@). > >> > >> > >> - Refactoring 2: Refactoring 2 is an improved version of the > >> refactorings developed by Gustavos Santos and they should used to > >> replace the existing one. > >> > >> > >> ## VM > >> > >> - 64-bits by default: Mac and Linux 64 bits VMs have been working for > >> over a year and since June 2017 a Windows 64 bits VM has been working. > >> The plan is to stabilize each part of Pharo that is not working in 64 > >> bits as well as in 32 bits and make the 64 bits Pharo images and VM > >> the default download for Pharo. > >> > >> - Headless VM: Ronie Salgado has built a headless VM on his VM branch, > >> we need to merge and check everything works fine. With the headless > >> VM, Pharo can be run in true headless mode (not with a hidden window) > >> and it is required for future VM features (embeddable VM, ...). > >> > >> - Embeddable VM: The VM should be able to be embedded in external > >> applications, the application accessing objects through well-defined > >> APIs. > >> > >> - Idle VM: The goal is to avoid the active waiting loop consuming > >> several per cent of cpu time when the Pharo image is not doing > >> anything. > >> > >> - Android VM: integration of the Android VM build and tests in the > >> integration setup. > >> > >> - Threaded FFI: all FFI calls should be non-blocking (reviving > >> prototype of Eliot) > >> > >> - ZeroConf for ARM: The ARM VM should be available from the zeroConf servers > >> > >> - Better support for large heaps (GC tuning API, incremental GC). > >> Pharo 64 bit is now able to manage large heap. However better > >> performance can be proposed by offering better settings for the > >> different GC zone. > > > > The most important thing here is the incremental GC. Spur has a generation scavenger that collects garbage in newly created objects (new space), and a mark-compact collector that collects and compacts garbage in old space. > > > > Right now on my 2.3GHz MacMini doing normal development, the generation scavenger causes pauses of 1ms or less, and the mark-compact collector than causes pauses of around 200ms. Both account for about 0.75% of entire execution time (1.5% total), so the scavenger is invoked frequently and the pauses that it creates are not noticeable. But the occasional pauses created by the mark-compact collector /are/ noticeable, especially in games and music. > > > > The idea for the incremental collector is to implement a mark-sweep or mark-sweep-compact collector for old space that works incrementally, doing a little bit of work on each invocation, probably immediately after a scavenge. It is intended to avoid the long pauses caused by the non-incremental mark-compact collector and make the system more suitable for games, music, etc. > > > > [Work on scaling for large 64-bit heaps is somewhere in the future, but right now we don't have any applications like this generating demand for a solution.] > > Actually, being able to efficiently keep a couple of GB worth of objects in memory, even better 10s of GB seems like an important short term goal. I for one don't really care for games or music. I do care about server software and there we need to use the memory that is available so cheaply today. Luckily we now have 64-bit support. > > Indeed, and the reason I designed Spur to be both 32-bit and 64-bit was because my employer Cadence wanted to scale to much larger models (the application manipulates large models of SystemOnAChip (SOC) components expressed in XML). But their application is batch and they're happy just with the ability to go somewhat above the approximately 3 Gb limit of the 32-bit system. > > But what I was getting at in my comment was that, while 64-bit server applications are indeed very important, I have not been contacted by anyone yet complaining about scaling problems with the 64-bit system. This could be for a number of reasons, including that there are not many large applications out there. We know that pause times in the full GC are an issue for animation ad music software. I didn't articulate it, but I also think that an incremental GC provides somewhat of an answer to scaling for 64-bit servers. > > Let's say that currently the mark-compact collector collects at about 1Gb per second (that's about what I'm seeing on my MacMini for a 600Mb heap, collected and compacted in about 530ms on a 2.3GHz Core i7, = 1.1Gb/s). For the moment let's assume the GC scales linearly. Then a 10Gb server application will occasionally suffer 10 second pauses, which is unacceptable for any kind of high-availability server. The total execution time overhead of GC may be in the 1%-10% range, but the pause times may be unacceptable (note that they're /not/ unacceptable for my employer's batch usage). > > So were the GC to be extended with an incremental collector we may be able to bring pause times back down to the tens of milliseconds rage or below. Clearly how the server uses memory, and how the GC scales are crucially important. Right now I have no data on either; no large 64-bit server applications to analyze and consequently not much idea of how the current GC or an incremental GC will scale. So for the moment I'm going to prioritize doing an incremental GC, and making it measurable and tunable. Hopefully the incremental GC will benefit both large server applications as well as music and animation. > > But nothing will substitute for experience with real applications. So far there are very few large 64-bit applications running on Cog for me to analyze. Even Cadence's customers are too protective of their data for me to analyze real applications. Thanks for the reply, Eliot. Maybe is was (up until we got 64-bit for real) a chicken-egg problem: it was not realistic to build big applications due to the limits of 32-bit images. I am sure this will change in the future. Sven > >> - Integrate various fixes to support better high resolution display. > >> In Retina mac display Pharo looks blurry. The fix to support Retina > >> should be integrated in the VM. > >> > >> ### Sista-related > >> Sista is an optimizing JIT for Pharo. It is the result of multiple > >> years of developement by Clement Bera and Eliot Miranda. An optimizing > >> JIT is manipulating code (folding constant, rewriting it) before > >> compiling it to assembly code. > > > > Actually this is a speculative inliner (also known as an adaptive optimizer). It works by creating special optimized versions of methods that inline smaller methods, thereby allowing optimising away unnecessary tests, sends and blocks, etc. it is driven by "hot spot" detectors in the first-level JIT code and so inlines code that is better my used a lot. It is quite similar to the optimisers in Java HotSpot, JavaScript V8, etc. its implementation and chitecture is more like Graal, because the inliner, optimiser and deoptimiser are all in Smalltalk up in the image. This in-image optimizer is called Scorch. The overall architecture is called Sista, for Speculative Inlining Smalltalk Architecture. > > > >> - New Block Closure implementation by default: Allows one to implement > >> in the Opal compiler the copying and clean blocks optimisations, > >> reduce massively the complexity of some parts of the JIT (debugging > >> API to map machine code pc to bytecode pc for example) and is required > >> for the Sista support. Some work remains in debugging/IDE support. > >> > >> - New Bytecode set in production: Eases bytecode decoding (simplifying > >> the code base of the bytecode to source code decompiler, the debugger, > >> the JIT, etc.), lifts compiled code limitations (size of jumps, etc.) > >> and is required for the Sista support. Some work remains in debugging > >> support. > >> > >> - Read-only objects: They work in Pharo 6, but the in-image support > >> needs to be improved (fall-back code of primitives, etc.) and some > >> polishing needs to be done (primitive error code not always correct, > >> etc.). Used for the support of different project, including GBS. > > > >> - Literals immutability (based on read-only objects). In particular an > >> analysis of the image usage of literal (string mutation) is needed. > > > >> - Sista preview: A first version of Sista will be integrated, yielding > >> 1.5x performance boost on most applications, but will be optional and > >> will require specific constraints (not toying too much with literal > >> mutability, partial IDE support, etc.). > >> > >> > >> ### Already done > >> > >> - [DONE] Remove Shoreline reporter. > >> - [DONE] Kernel should not depend on Traits: This will speed up the > >> bootstrap and support the modular introduction of alternate traits > >> implementation currently designed by Pablo Tesone and tested in the > >> new generation of Moose. > > > > Eliot > > _,,,^..^,,,_ (phone) > > > > > > -- > _,,,^..^,,,_ > best, Eliot

July 6, 2017

Re: [Pharo-dev] Discussing the roadmap

by Eliot Miranda

Hi Sven, On Thu, Jul 6, 2017 at 10:10 AM, Sven Van Caekenberghe <sven(a)stfx.eu> wrote: > > > On 6 Jul 2017, at 18:41, Eliot Miranda <eliot.miranda(a)gmail.com> wrote: > > > > Hi Stef, > > > >> On Jul 6, 2017, at 1:55 AM, Stephane Ducasse <stepharo.self(a)gmail.com> > wrote: > >> > >> We would like to share this list with you and get your feedback and > inputs. > >> It will be presented and discussed again at ESUG. > >> > >> Stef on the behalf of the engineering team of the Pharo consortium. > >> > >> # Pharo 7 (and 8) potential roadmap > >> > >> This document contains a list of actions that should be done during > >> Pharo 7 and 8. > >> It is not a definitive list. It is the result of a discussion within > >> the engineer team and RModers. > >> It first lists actions that should be performed at the image level > >> then lists the actions for the VM. > >> > >> > >> ## Image > >> As a general principle, we will try to remove something when we add a > >> new features or version of a component. > >> > >> ### New tools > >> - Iceberg: Iceberg is the new tool suite to handle new VCS in Pharo. > >> For the moment it supports Git. This tool will become central to the > >> development of projects in Pharo. The key and first enhancements will > >> be: > >> - cherry picking > >> - multiple directories support, subtrees > >> - better new development process support > >> > >> - Calypso: Calypso is a new tool suite for editing and navigating > >> code. It is modular and can easily be extended. > >> - integration, set as the default browser > >> > >> - Hermes (binary loader): Hermes is a code loader that does not > >> require the compiler to be loaded. It will be used to speed up the > >> bootstrap. > >> > >> - Beacon: Beacon is an announcement-based logger. It should be > >> introduced but in addition the left over of previous logging should be > >> cleaned and removed for example many of the transcript show: should be > >> removed. > >> > >> - Inspector refreshing: the inspector should refresh its values by > default. > >> > >> - Cargo: Cargo is a new package manager. It supports the expression of > >> dependencies between packages. We are currently validating that it can > >> support conditional loading (platform) and building new tooling to > >> express and validate data. > >> > >> - Check dependencies when committing. Pharo comes with a dependency > >> analyser tools. Such tools should be used before commiting to warn the > >> user when a new dependency is introduced in a package. > >> > >> ### Cleaning bloat > >> - Removing of Nautilus: Once Calypso will be integrated and exhibit > >> similar fetaures than Nautilus. Nautilus should be removed. > >> - Remove old text editor: there is only one or two widgets still using > >> the old text editor and the rubric text editor. > >> - Remove TxText: TxText was an attempt to get a new generation text > >> editor. Now it has been rewritten with a new design in Bloc so we > >> should remove it since Bloc editor is better and actively maintained. > >> - Remove Komitter: Iceberg already supports cherrypicking on commit > >> therefore Komitter can be safely removed from Pharo. > >> - Remove system categorizer: The old system categorizer is not used > >> anymore and should be remove. > >> > >> - Old compiler removal: The old compiler should now get unloaded from > Pharo. > >> - The old compiler should be moved to an external package and make > >> sure that it can be reloaded. > >> - In addition the encoders should be separated. (@@ more here@@) > >> > >> - Old inspector cleanup: we should remove the eyeInspector framework > >> but we should introduce a minimal fallback inspector. This inspector > >> should work without Spec or any frameworks. > >> > >> - Clean behavior protocol. The number of methods in the core Behavior, > >> ClassDescription and Class requires some analysis and cleaning. > >> > >> - Kernel modularization: the effort on modularising the packages > >> should be continued. > >> > >> ### Language infrastructure > >> The following points are more related to the infrastructure of > >> manipulating and loading class definitions. There are the basis for > >> the future module system and cleaning some often hidden parts of the > >> system. > >> > >> - New class definition: The class definition is not scaling anymore > >> due to the large number of options (traits, slot, tage, package, > >> immediate, ephemeron). A fluid-based message API should be designed. > >> > >> - Support for Undefined classes: When loading old code or code whose > >> superclass has not yet being loaded, the system has inconsistent > >> behavior. Depending on the tools, it may not load the code, raise a > >> warning or decide that the superclass is Object without any other > >> notice and losing the name of the original superclass. We are working > >> on a new mechanism to support a unique way to handle such case. To be > >> presented at IWST/ESUG > >> > >> - Class definition parser: Class definition is parser in different > >> place of the system and in addition the output is the direct creation > >> of a class object instead of an object representing the class > >> definition that can be further manipulated. We are working on class > >> definition parser. It produces a separate AST. It will help the > >> building of tools. > >> > >> - Better update infrastructure. Pablo Tesone has been working on a > >> better update mechanism, better modular class builder. > >> > >> - Ring2: Ring is a metamodel to manipulate code that is not actively > >> running in the current system. It is useful to perform analysis (such > >> as browsing, navigating off-line or remote definitions) before > >> actually loading the code. Ring got redesigned by Pavel Krivanek to > >> support the bootstrap of Pharo 60. The results is Ring2. > >> > >> - Opal is the Pharo Compiler. It should be enhanced to support handle > >> better the warning and a general enhancement pass is needed. > >> > >> ### System enhancements > >> > >> - Commandline enhancements. RMOD is currently improving the > >> command-line infrastructure and making sure that the system can work > >> in read-only mode. > >> > >> - Cleaning streams: the idea is to make sure that the system does not > >> use the old streams. The idea is to start using the fileStreams and > >> make sure that the Stream package can be substituated in the future by > >> other streams with the same API. Therefore hardcoded class names > >> should be substituted with factory (readSream writeStream). From that > >> perspective we do not think that it is wise to introduce Xtreams. We > >> should analyse the API of the current implementation. > >> > >> - New theme from the beginning. It is really important that each > >> version of Pharo is identified open the default opening. Pharo should > >> come up with two default themes: one light and one dark. > >> > >> - Better themes/palettes support > >> - Better and nicer Tabs. Tabs design should be revisited. > >> > >> - OSWindow world rendering: the effort to remove the logic of the > >> windowing from the VM should be continued. OSWindow should be used > >> instead. > >> > >> - Freetype: The current freetype plugin is the source of many bugs and > >> problem. Thibault Raffaillac used FFI to do direct call to openGL > >> (@@to verify@@). > >> > >> > >> - Refactoring 2: Refactoring 2 is an improved version of the > >> refactorings developed by Gustavos Santos and they should used to > >> replace the existing one. > >> > >> > >> ## VM > >> > >> - 64-bits by default: Mac and Linux 64 bits VMs have been working for > >> over a year and since June 2017 a Windows 64 bits VM has been working. > >> The plan is to stabilize each part of Pharo that is not working in 64 > >> bits as well as in 32 bits and make the 64 bits Pharo images and VM > >> the default download for Pharo. > >> > >> - Headless VM: Ronie Salgado has built a headless VM on his VM branch, > >> we need to merge and check everything works fine. With the headless > >> VM, Pharo can be run in true headless mode (not with a hidden window) > >> and it is required for future VM features (embeddable VM, ...). > >> > >> - Embeddable VM: The VM should be able to be embedded in external > >> applications, the application accessing objects through well-defined > >> APIs. > >> > >> - Idle VM: The goal is to avoid the active waiting loop consuming > >> several per cent of cpu time when the Pharo image is not doing > >> anything. > >> > >> - Android VM: integration of the Android VM build and tests in the > >> integration setup. > >> > >> - Threaded FFI: all FFI calls should be non-blocking (reviving > >> prototype of Eliot) > >> > >> - ZeroConf for ARM: The ARM VM should be available from the zeroConf > servers > >> > >> - Better support for large heaps (GC tuning API, incremental GC). > >> Pharo 64 bit is now able to manage large heap. However better > >> performance can be proposed by offering better settings for the > >> different GC zone. > > > > The most important thing here is the incremental GC. Spur has a > generation scavenger that collects garbage in newly created objects (new > space), and a mark-compact collector that collects and compacts garbage in > old space. > > > > Right now on my 2.3GHz MacMini doing normal development, the generation > scavenger causes pauses of 1ms or less, and the mark-compact collector than > causes pauses of around 200ms. Both account for about 0.75% of entire > execution time (1.5% total), so the scavenger is invoked frequently and the > pauses that it creates are not noticeable. But the occasional pauses > created by the mark-compact collector /are/ noticeable, especially in games > and music. > > > > The idea for the incremental collector is to implement a mark-sweep or > mark-sweep-compact collector for old space that works incrementally, doing > a little bit of work on each invocation, probably immediately after a > scavenge. It is intended to avoid the long pauses caused by the > non-incremental mark-compact collector and make the system more suitable > for games, music, etc. > > > > [Work on scaling for large 64-bit heaps is somewhere in the future, but > right now we don't have any applications like this generating demand for a > solution.] > > Actually, being able to efficiently keep a couple of GB worth of objects > in memory, even better 10s of GB seems like an important short term goal. I > for one don't really care for games or music. I do care about server > software and there we need to use the memory that is available so cheaply > today. Luckily we now have 64-bit support. > Indeed, and the reason I designed Spur to be both 32-bit and 64-bit was because my employer Cadence wanted to scale to much larger models (the application manipulates large models of SystemOnAChip (SOC) components expressed in XML). But their application is batch and they're happy just with the ability to go somewhat above the approximately 3 Gb limit of the 32-bit system. But what I was getting at in my comment was that, while 64-bit server applications are indeed very important, I have not been contacted by anyone yet complaining about scaling problems with the 64-bit system. This could be for a number of reasons, including that there are not many large applications out there. We know that pause times in the full GC are an issue for animation ad music software. I didn't articulate it, but I also think that an incremental GC provides somewhat of an answer to scaling for 64-bit servers. Let's say that currently the mark-compact collector collects at about 1Gb per second (that's about what I'm seeing on my MacMini for a 600Mb heap, collected and compacted in about 530ms on a 2.3GHz Core i7, = 1.1Gb/s). For the moment let's assume the GC scales linearly. Then a 10Gb server application will occasionally suffer 10 second pauses, which is unacceptable for any kind of high-availability server. The total execution time overhead of GC may be in the 1%-10% range, but the pause times may be unacceptable (note that they're /not/ unacceptable for my employer's batch usage). So were the GC to be extended with an incremental collector we may be able to bring pause times back down to the tens of milliseconds rage or below. Clearly how the server uses memory, and how the GC scales are crucially important. Right now I have no data on either; no large 64-bit server applications to analyze and consequently not much idea of how the current GC or an incremental GC will scale. So for the moment I'm going to prioritize doing an incremental GC, and making it measurable and tunable. Hopefully the incremental GC will benefit both large server applications as well as music and animation. But nothing will substitute for experience with real applications. So far there are very few large 64-bit applications running on Cog for me to analyze. Even Cadence's customers are too protective of their data for me to analyze real applications. >> - Integrate various fixes to support better high resolution display. > >> In Retina mac display Pharo looks blurry. The fix to support Retina > >> should be integrated in the VM. > >> > >> ### Sista-related > >> Sista is an optimizing JIT for Pharo. It is the result of multiple > >> years of developement by Clement Bera and Eliot Miranda. An optimizing > >> JIT is manipulating code (folding constant, rewriting it) before > >> compiling it to assembly code. > > > > Actually this is a speculative inliner (also known as an adaptive > optimizer). It works by creating special optimized versions of methods > that inline smaller methods, thereby allowing optimising away unnecessary > tests, sends and blocks, etc. it is driven by "hot spot" detectors in the > first-level JIT code and so inlines code that is better my used a lot. It > is quite similar to the optimisers in Java HotSpot, JavaScript V8, etc. > its implementation and chitecture is more like Graal, because the inliner, > optimiser and deoptimiser are all in Smalltalk up in the image. This > in-image optimizer is called Scorch. The overall architecture is called > Sista, for Speculative Inlining Smalltalk Architecture. > > > >> - New Block Closure implementation by default: Allows one to implement > >> in the Opal compiler the copying and clean blocks optimisations, > >> reduce massively the complexity of some parts of the JIT (debugging > >> API to map machine code pc to bytecode pc for example) and is required > >> for the Sista support. Some work remains in debugging/IDE support. > >> > >> - New Bytecode set in production: Eases bytecode decoding (simplifying > >> the code base of the bytecode to source code decompiler, the debugger, > >> the JIT, etc.), lifts compiled code limitations (size of jumps, etc.) > >> and is required for the Sista support. Some work remains in debugging > >> support. > >> > >> - Read-only objects: They work in Pharo 6, but the in-image support > >> needs to be improved (fall-back code of primitives, etc.) and some > >> polishing needs to be done (primitive error code not always correct, > >> etc.). Used for the support of different project, including GBS. > > > >> - Literals immutability (based on read-only objects). In particular an > >> analysis of the image usage of literal (string mutation) is needed. > > > >> - Sista preview: A first version of Sista will be integrated, yielding > >> 1.5x performance boost on most applications, but will be optional and > >> will require specific constraints (not toying too much with literal > >> mutability, partial IDE support, etc.). > >> > >> > >> ### Already done > >> > >> - [DONE] Remove Shoreline reporter. > >> - [DONE] Kernel should not depend on Traits: This will speed up the > >> bootstrap and support the modular introduction of alternate traits > >> implementation currently designed by Pablo Tesone and tested in the > >> new generation of Moose. > > > > Eliot > > _,,,^..^,,,_ (phone) > > > -- _,,,^..^,,,_ best, Eliot

July 6, 2017

Re: [Pharo-dev] Discussing the roadmap

by Eliot Miranda

Hi Stef, On Thu, Jul 6, 2017 at 1:20 PM, Stephane Ducasse <stepharo.self(a)gmail.com> wrote: > I updated the roadmap. > > Btw eliot I think that you should sync with clement and esteban. > trust me, we do. > I'm not the guy taking decision here. We should agree on the key > elements of the roadmap > as well as requirements expressed by people that put money on the table. > Once all the agenda/requirements/analyse is done then the consortium > can take a decision for the > benefit of all. Now about large images, some companies really need that. > Indeed. See my reply to Sven. > How much time do you think that an incremental GC implementation would > take? > Three months. > > Stef > > On Thu, Jul 6, 2017 at 6:41 PM, Eliot Miranda <eliot.miranda(a)gmail.com> > wrote: > > Hi Stef, > > > >> On Jul 6, 2017, at 1:55 AM, Stephane Ducasse <stepharo.self(a)gmail.com> > wrote: > >> > >> We would like to share this list with you and get your feedback and > inputs. > >> It will be presented and discussed again at ESUG. > >> > >> Stef on the behalf of the engineering team of the Pharo consortium. > >> > >> # Pharo 7 (and 8) potential roadmap > >> > >> This document contains a list of actions that should be done during > >> Pharo 7 and 8. > >> It is not a definitive list. It is the result of a discussion within > >> the engineer team and RModers. > >> It first lists actions that should be performed at the image level > >> then lists the actions for the VM. > >> > >> > >> ## Image > >> As a general principle, we will try to remove something when we add a > >> new features or version of a component. > >> > >> ### New tools > >> - Iceberg: Iceberg is the new tool suite to handle new VCS in Pharo. > >> For the moment it supports Git. This tool will become central to the > >> development of projects in Pharo. The key and first enhancements will > >> be: > >> - cherry picking > >> - multiple directories support, subtrees > >> - better new development process support > >> > >> - Calypso: Calypso is a new tool suite for editing and navigating > >> code. It is modular and can easily be extended. > >> - integration, set as the default browser > >> > >> - Hermes (binary loader): Hermes is a code loader that does not > >> require the compiler to be loaded. It will be used to speed up the > >> bootstrap. > >> > >> - Beacon: Beacon is an announcement-based logger. It should be > >> introduced but in addition the left over of previous logging should be > >> cleaned and removed for example many of the transcript show: should be > >> removed. > >> > >> - Inspector refreshing: the inspector should refresh its values by > default. > >> > >> - Cargo: Cargo is a new package manager. It supports the expression of > >> dependencies between packages. We are currently validating that it can > >> support conditional loading (platform) and building new tooling to > >> express and validate data. > >> > >> - Check dependencies when committing. Pharo comes with a dependency > >> analyser tools. Such tools should be used before commiting to warn the > >> user when a new dependency is introduced in a package. > >> > >> ### Cleaning bloat > >> - Removing of Nautilus: Once Calypso will be integrated and exhibit > >> similar fetaures than Nautilus. Nautilus should be removed. > >> - Remove old text editor: there is only one or two widgets still using > >> the old text editor and the rubric text editor. > >> - Remove TxText: TxText was an attempt to get a new generation text > >> editor. Now it has been rewritten with a new design in Bloc so we > >> should remove it since Bloc editor is better and actively maintained. > >> - Remove Komitter: Iceberg already supports cherrypicking on commit > >> therefore Komitter can be safely removed from Pharo. > >> - Remove system categorizer: The old system categorizer is not used > >> anymore and should be remove. > >> > >> - Old compiler removal: The old compiler should now get unloaded from > Pharo. > >> - The old compiler should be moved to an external package and make > >> sure that it can be reloaded. > >> - In addition the encoders should be separated. (@@ more here@@) > >> > >> - Old inspector cleanup: we should remove the eyeInspector framework > >> but we should introduce a minimal fallback inspector. This inspector > >> should work without Spec or any frameworks. > >> > >> - Clean behavior protocol. The number of methods in the core Behavior, > >> ClassDescription and Class requires some analysis and cleaning. > >> > >> - Kernel modularization: the effort on modularising the packages > >> should be continued. > >> > >> ### Language infrastructure > >> The following points are more related to the infrastructure of > >> manipulating and loading class definitions. There are the basis for > >> the future module system and cleaning some often hidden parts of the > >> system. > >> > >> - New class definition: The class definition is not scaling anymore > >> due to the large number of options (traits, slot, tage, package, > >> immediate, ephemeron). A fluid-based message API should be designed. > >> > >> - Support for Undefined classes: When loading old code or code whose > >> superclass has not yet being loaded, the system has inconsistent > >> behavior. Depending on the tools, it may not load the code, raise a > >> warning or decide that the superclass is Object without any other > >> notice and losing the name of the original superclass. We are working > >> on a new mechanism to support a unique way to handle such case. To be > >> presented at IWST/ESUG > >> > >> - Class definition parser: Class definition is parser in different > >> place of the system and in addition the output is the direct creation > >> of a class object instead of an object representing the class > >> definition that can be further manipulated. We are working on class > >> definition parser. It produces a separate AST. It will help the > >> building of tools. > >> > >> - Better update infrastructure. Pablo Tesone has been working on a > >> better update mechanism, better modular class builder. > >> > >> - Ring2: Ring is a metamodel to manipulate code that is not actively > >> running in the current system. It is useful to perform analysis (such > >> as browsing, navigating off-line or remote definitions) before > >> actually loading the code. Ring got redesigned by Pavel Krivanek to > >> support the bootstrap of Pharo 60. The results is Ring2. > >> > >> - Opal is the Pharo Compiler. It should be enhanced to support handle > >> better the warning and a general enhancement pass is needed. > >> > >> ### System enhancements > >> > >> - Commandline enhancements. RMOD is currently improving the > >> command-line infrastructure and making sure that the system can work > >> in read-only mode. > >> > >> - Cleaning streams: the idea is to make sure that the system does not > >> use the old streams. The idea is to start using the fileStreams and > >> make sure that the Stream package can be substituated in the future by > >> other streams with the same API. Therefore hardcoded class names > >> should be substituted with factory (readSream writeStream). From that > >> perspective we do not think that it is wise to introduce Xtreams. We > >> should analyse the API of the current implementation. > >> > >> - New theme from the beginning. It is really important that each > >> version of Pharo is identified open the default opening. Pharo should > >> come up with two default themes: one light and one dark. > >> > >> - Better themes/palettes support > >> - Better and nicer Tabs. Tabs design should be revisited. > >> > >> - OSWindow world rendering: the effort to remove the logic of the > >> windowing from the VM should be continued. OSWindow should be used > >> instead. > >> > >> - Freetype: The current freetype plugin is the source of many bugs and > >> problem. Thibault Raffaillac used FFI to do direct call to openGL > >> (@@to verify@@). > >> > >> > >> - Refactoring 2: Refactoring 2 is an improved version of the > >> refactorings developed by Gustavos Santos and they should used to > >> replace the existing one. > >> > >> > >> ## VM > >> > >> - 64-bits by default: Mac and Linux 64 bits VMs have been working for > >> over a year and since June 2017 a Windows 64 bits VM has been working. > >> The plan is to stabilize each part of Pharo that is not working in 64 > >> bits as well as in 32 bits and make the 64 bits Pharo images and VM > >> the default download for Pharo. > >> > >> - Headless VM: Ronie Salgado has built a headless VM on his VM branch, > >> we need to merge and check everything works fine. With the headless > >> VM, Pharo can be run in true headless mode (not with a hidden window) > >> and it is required for future VM features (embeddable VM, ...). > >> > >> - Embeddable VM: The VM should be able to be embedded in external > >> applications, the application accessing objects through well-defined > >> APIs. > >> > >> - Idle VM: The goal is to avoid the active waiting loop consuming > >> several per cent of cpu time when the Pharo image is not doing > >> anything. > >> > >> - Android VM: integration of the Android VM build and tests in the > >> integration setup. > >> > >> - Threaded FFI: all FFI calls should be non-blocking (reviving > >> prototype of Eliot) > >> > >> - ZeroConf for ARM: The ARM VM should be available from the zeroConf > servers > >> > >> - Better support for large heaps (GC tuning API, incremental GC). > >> Pharo 64 bit is now able to manage large heap. However better > >> performance can be proposed by offering better settings for the > >> different GC zone. > > > > The most important thing here is the incremental GC. Spur has a > generation scavenger that collects garbage in newly created objects (new > space), and a mark-compact collector that collects and compacts garbage in > old space. > > > > Right now on my 2.3GHz MacMini doing normal development, the generation > scavenger causes pauses of 1ms or less, and the mark-compact collector than > causes pauses of around 200ms. Both account for about 0.75% of entire > execution time (1.5% total), so the scavenger is invoked frequently and the > pauses that it creates are not noticeable. But the occasional pauses > created by the mark-compact collector /are/ noticeable, especially in games > and music. > > > > The idea for the incremental collector is to implement a mark-sweep or > mark-sweep-compact collector for old space that works incrementally, doing > a little bit of work on each invocation, probably immediately after a > scavenge. It is intended to avoid the long pauses caused by the > non-incremental mark-compact collector and make the system more suitable > for games, music, etc. > > > > [Work on scaling for large 64-bit heaps is somewhere in the future, but > right now we don't have any applications like this generating demand for a > solution.] > > > > > >> - Integrate various fixes to support better high resolution display. > >> In Retina mac display Pharo looks blurry. The fix to support Retina > >> should be integrated in the VM. > >> > >> ### Sista-related > >> Sista is an optimizing JIT for Pharo. It is the result of multiple > >> years of developement by Clement Bera and Eliot Miranda. An optimizing > >> JIT is manipulating code (folding constant, rewriting it) before > >> compiling it to assembly code. > > > > Actually this is a speculative inliner (also known as an adaptive > optimizer). It works by creating special optimized versions of methods > that inline smaller methods, thereby allowing optimising away unnecessary > tests, sends and blocks, etc. it is driven by "hot spot" detectors in the > first-level JIT code and so inlines code that is better my used a lot. It > is quite similar to the optimisers in Java HotSpot, JavaScript V8, etc. > its implementation and chitecture is more like Graal, because the inliner, > optimiser and deoptimiser are all in Smalltalk up in the image. This > in-image optimizer is called Scorch. The overall architecture is called > Sista, for Speculative Inlining Smalltalk Architecture. > > > >> - New Block Closure implementation by default: Allows one to implement > >> in the Opal compiler the copying and clean blocks optimisations, > >> reduce massively the complexity of some parts of the JIT (debugging > >> API to map machine code pc to bytecode pc for example) and is required > >> for the Sista support. Some work remains in debugging/IDE support. > >> > >> - New Bytecode set in production: Eases bytecode decoding (simplifying > >> the code base of the bytecode to source code decompiler, the debugger, > >> the JIT, etc.), lifts compiled code limitations (size of jumps, etc.) > >> and is required for the Sista support. Some work remains in debugging > >> support. > >> > >> - Read-only objects: They work in Pharo 6, but the in-image support > >> needs to be improved (fall-back code of primitives, etc.) and some > >> polishing needs to be done (primitive error code not always correct, > >> etc.). Used for the support of different project, including GBS. > > > >> - Literals immutability (based on read-only objects). In particular an > >> analysis of the image usage of literal (string mutation) is needed. > > > >> - Sista preview: A first version of Sista will be integrated, yielding > >> 1.5x performance boost on most applications, but will be optional and > >> will require specific constraints (not toying too much with literal > >> mutability, partial IDE support, etc.). > >> > >> > >> ### Already done > >> > >> - [DONE] Remove Shoreline reporter. > >> - [DONE] Kernel should not depend on Traits: This will speed up the > >> bootstrap and support the modular introduction of alternate traits > >> implementation currently designed by Pablo Tesone and tested in the > >> new generation of Moose. > > > > Eliot > > _,,,^..^,,,_ (phone) > > -- _,,,^..^,,,_ best, Eliot

July 6, 2017

Jump to page: