I am the only potential customer of CPPBridge because a) I am the only C++ coder AFAIK b) Using DLLs (dynamically linked shared libraries) is still by far the best solution because using the UFFI you can do everything from Pharo. Compared to CPPBridge which you will need to code in C++.
So as you can imagine making a general library for IPC communication via shared memory is not my goal. The Unreal part will be a separate library that will depend on this one. I have not touched that one yet but I will soon. UnrealBridge library will become my most important Pharo project because the goal is to start selling games using Pharo and Unreal. IPC Protocol wise , I don't think synchronisation will be an issue because Unreal will be too fast for Pharo anyway and its ok if Pharo lags behind even a few dozens of milliseconds which is a matter of losing a couple of frames. I may implement a hybrid synchronisation for when Pharo lag becomes too severe but that will have and see in practice. Something that interest me is "stealing" your idea for NBOpenGL where you made an automatic generator for the library that used OpenGL headers. This is something that could benefit me a lot because Unreal contains over 10.000 methods and wrapping them one by one is not something I am looking forward to doing. This means generating both C++ and Pharo code. Of course if I feel that what I implement with UnrealBridge is general purpose enough I can port it back to CPPBridge. In any case I must say it was a lot of fun accomplishing this goal and cant wait to dive deeper into IPC magic, making Pharo the nerve centre, the brain , of my coding environment is a role that I think Pharo would excel at. It will also allow me to use Pharo as much as possible and provide a super powerful game / graphics engine to the Pharo community. Shared memory is that just the means to that goal. Moose could also a play role later on for analysing my game code both in Pharo and C++ and use Roassal for some nifty visualizations though I could use Unreal as backend to Roassal for some nice 3d visualisations. So ton of ideas, tons of fun and a ton of time to accomplish them. PS: I have to say you have been inspiration for my IPC saga, you did the unthinkable (at least to me) and brought Assembly to Pharo, I never had a use for Assembly but nonetheless you taught me that Pharo potential is more massive than I imagined, so thank you :) On Wed, Nov 9, 2016 at 3:21 AM Igor Stasenko <siguc...@gmail.com> wrote: > On 9 November 2016 at 02:09, Dimitris Chloupis <kilon.al...@gmail.com> > wrote: > > I have only played with c structs and it looks like they do the job fine . > If json parsing is slow especially in Pharo then it's out of the question. > Parsing must be kept close to zero because I will be running some very > tight loops of 100 frames per second and there is no way that I will let > Pharo take its time as the whole idea of using the fastest way to IPC was > to keep Pharo in sync with unreal . Heavy lifting will be done only by > Unreal. Pharo will be used just for logic. > > Well, it depends what you want: if you want to give away a > complete/general solution for IPC with Pharo via shared memory, then you > should create some kind of library (both C++ and Pharo) which would allow > users to setup the bridge configuration in the way they want, leaving the > application-specific exchange to the hands of users.. And if you just want > to make own data exchange for own purpose, then you are basically done. > > > In any case there is a ton of testing and profiling needed to be done . So > these are just the first steps. > On Wed, 9 Nov 2016 at 02:58, Igor Stasenko <siguc...@gmail.com> wrote: > > JSON? -- But you were talking about speed. Once you put JSON as a base for > IPC, then say goodbye to speed. > > Because JSON parsing/writing will eat like 90% of time even if you would > use sockets. > So, why not using some kind of binary protocol? > Unless you wanna use JSON for initial handshaking exchange. Then it is > quite reasonable. > > But all of that still won't free you from inventing some kind of > contract(s) which would allow parties to agree who writes where and/or who > writes, while other waits then reads etc. > > > On 9 November 2016 at 00:06, Dimitris Chloupis <kilon.al...@gmail.com> > wrote: > > > *https://youtu.be/pI4PR3XaX6Q <https://youtu.be/pI4PR3XaX6Q>* > > *What is it ?* > > CPPBridge is a library that allows Pharo to share memory with any C/C++ > application. Opening the door not only to IPC and data sharing but also > even complete control of C++ code from Pharo and vice versa. > > *How to install ?* > > In a few hours it should be available from Package Browser, if not you can > always fetch it with Metacello from here because it comes with a Baseline > > https://github.com/kilon/CPPBridge > > *Why bother making such a library ? * > > In my saga to find a way to use Pharo as a scripting language for Unreal > Game Engine, I had two options > > a) Build Unreal as a Library and use Pharo UFFI to launch and control it > b) Embed Pharo inside the Unreal Executable (this is what every other > scripting language uses for controlling Unreal) > > Option (a) was a no go, because Unreal is huge , complex and uses its own > custom made build tools, making a DLL for Pharo or an army of DLLs out of > the question > > Option (b) Embeding Pharo inside an executable is impossible and > implementing it also insanely complex > > Naturally my mind went first into sockets which is what I use to make > Pharo able to use Python libraries. However sockets have proven too slow > for the insanely fast loops of Unreal. > > *What are the advantages ?* > > 1) *No need to move data around.* Sharing memory means you don't have to > move data around, you can use directly the shared memory > > 2)* Extend the Pharo image beyond Pharo.* Shared memory is mapped into a > file means that you can do with C++ what you can do with Pharo image, save > the live state directly to a binary file. That means we no longer need to > worry about sessions and reinitializing C/C++ data since memory mapped file > acts as an extension of the Pharo image. > > 3) *Blazing fast. *Memory mapping is a function that comes directly from > the OS kernel and as such it has to be extremely fast. Memory mapping is > also what is used for dynamically linked shared libraries an extremely > important feature for any application including Pharo that heavily depends > on (see Cairo for Athens). So its a very mature , well tested method. > > 4) *No extra libraries needed* to be installed, CPPBridge uses OS > libraries to work out of the box > > 5) *Low level handling of memory.* Direct access to memory you can even > manipulate the memory byte by byte > > 6)* Memory efficient.* Memory mapping excels at large data, the bigger > the better. Can take full advantage of your entire free memory and not > waste a byte. That means also that can be used to optimise Pharo memory, > since you could compress your Pharo objects to bytes and mapped file will > store the live state. > > 7) *Tons of Languages. *Because memory mapping is a core functionality > for every OS out there, pretty much every programming language supports it. > CPPBridge currently supports only C/C++ but all languages can be supported > giving access to Pharo to any library for any programming language. Sky is > the limit > > 8) *Self Documented. *CPPBridge is small, simple and with large class > comment and comments for every method. YouTube video tutorial also > available and linked on top. > > 9) *Works both ways*, C/C++ and Pharo can access and modify the shared > memory. Making it possible for C/C++ to use Pharo libraries and Pharo to > use C/C++ libraries. > > 10) Experiments have proven that it improves sex life... if it does not > please file a bug report ;) > > > *What are the disadvantages ?* > > 1) *Candy Crash Saga*. Dare do something incorrectly and Pharo will > crash. CPPBridge can easily point to wrong address if you are not aware of > what you doing. > > 2) *C++/C* . If you think you can avoid learning C/C++ and that this is a > magic solution , think again. The C/C++ application must be modified to > include shared memory mapping for CPPBridge to work . > > 3) *Local only*. Unlike sockets, Shared Memory works only on the same > machine so no remote execution and manipulation of code like in my socket > bridge to Python > > 4) *UFFI still No1 option*. No replacement for UFFI it actually depends > on it to work , so if you can turn your C/C++ code into a DLL that should > be your first option. > > *Roadmap * > > Currently CPPBridge works only on MacOS , most likely on Linux too > (because it uses the Unix architecture) but I will have to test it. > > Windows is coming next ASAP, since its my No1 platform for creating > commercial games. > > Maybe establish a small protocol of communication via the Shared Memory , > JSON looks like a good universal format > > *Thanks* > > Big thanks to Eliot for inspiring me and Esteban for helping me figure out > things. > > > > > -- > Best regards, > Igor Stasenko. > > > > > -- > Best regards, > Igor Stasenko. >
