Hi Eliot, We were having some discussions about those topics, with Esteban Lorenzano and Stephane Ducasse. Esteban was the one who told me about callbacks, so I wrote according to what I remembered about that. About portability, only x86 ia32 is supported, with some stubs for ARM and PPC in the old FFI. In Pharo, only NativeBoost FFI is actually supported, at least in the official releases, in there end-user friendly callbacks are not supported. The main problem with adding those implementations is lack of good documentations. The C files of the old FFIs are poorly written, they have lots of #ifdef/#endif which makes reading their code really hard, except for the person who wrote them. Those C files should have a clear defined external interface and proper separations for architecture and operating systems. That is a problem because nobody wants to touch that code. About LLVM, this is only for FFI when you don't have a NativeBoost implementation for your platform, but there is already a LLVM JIT implemented, for example x86 ia32 and AMD64, ARM, probably PPC. You cannot compare our workforce with the one in LLVM and I think is a better idea to use it as a fallback for when we have nothing and then be able to add something better. For fast FFI you only have to create a C function that performs all the marshalling from the Smalltalk objects into the receiver C function types, then perform the actual C function call. That is the way that NativeBoost works, and it can be done by using LLVM, or even your favorite C compiler. So, about the bloating and dependency, I completely agree with you, but we need at least an interpreter based VM for x86 AMD64, ARM, PPC, and maybe some other popular/future platforms that supports communication with C. In fact, I am not interested in the LLVM optimizer, and probably I am not going to use it. I also know about the efforts on the virtual CPU, and I think that is the correct approach. The problem is how much time is going to take to complete for all the platforms, integrate it into NativeBoost and deploy it in the VM, with maybe at most two person at time working on them. In platforms, such as iOS you cannot use JIT at all, so you have to either use an interpreter or AOT everything and static link every dependency, that is where the static "fake FFI" enters into scene. BTW, end-user consumer applications in lot of platforms when deployed don't need or cannot have the full live coding system, debugger and browser, so losing a bit of flexibility here is not a problem. On January with Igor, we started making a direct graphics API, named OSWindow and we tossed all the vm-display code for Linux. If we continue with that work, FFI is going to become a core feature required to be able to make the main operating system Pharo window. So, if we don't have some portable way of doing FFI things in Pharo, being able to port it is going to become a lot more difficult. Greetings, Ronie Salgado 2014-02-13 17:22 GMT-03:00 Eliot Miranda <eliot.miranda@gmail.com>:
Hi Ronie,
On Thu, Feb 13, 2014 at 11:47 AM, Ronie Salgado <roniesalg@gmail.com>wrote:
I am proposing the following as a student:
Project idea
Name: Unified Foreign Function Interface Skill level: Advanced Possible Mentors: Igor Stasenko / Esteban Lorenzano Name of the Student: Ronie Salgado
Description:
Because NativeBoost has problems with portability and cannot be used anywhere(iOS), one proposal is making a modular unified foreign function interface. This UFFI should support multiples backends, a portable function calls specification, portable easy to use callbacks support, and portable foreign resources management.
Full Description Document: http://users.dcc.uchile.cl/~rsalgado/uffi-fourth-draft.pdf
This is interesting but I take issue with a couple of things. First of all
"The old Squeak FFI has problems with portability, callbacks and speed."
I disagree that there are problems with callbacks. In fact, IIRC the only callback facility we have is that which I implemented which is used both by Alien and FFI and works well enough that the Newspeak system used it to implement a native GUI on windows where callbacks are used to receive Windows events. That counts as working in my book. Also, the architecture I implemented is portable to other ABIs than x86.
But my major objection is to this:
"One interesting idea is to make a backend that uses LLVM compiler infrastruc- ture, because it is designed for compiling optimized C function and it already has multiples JIT backends. This small project is not to make a full Pharo VM JIT based in LLVM, because LLVM is not designed for that and it would require a lot of patching of LLVM."
IMO that's a very poor choice. LLVM is large, having a footprint of about a couple of megabytes. The Cog JIT is in contrast about 330k at its largest (Newspeak, support for two instruction sets, plus primitives, plus GC interface, etc, etc). If we as a community put effort into extending this we'll have a smaller footprint, a better JIT, and less dependencies. There's insignificant opportunity for using LLVM's optimizer when engineering call-outs and call-backs; they're too simple. So there's nothing to be gained other than bloat and dependency. And not putting effort into our own infrastructure is IMO a very bad idea.
Note the synergy with Clément and my work on speculative inlining which will extend the Cog JIT.
2014-02-13 12:21 GMT-03:00 tesonep@gmail.com <tesonep@gmail.com>:
Project idea
Description: Enhance refactorings with type information Skill level: Advance Possible Mentors: Guillermo Polito / Esteban Lorenzano / Santiago Bragagnolo Name of the Student: Pablo Tesone Keywords: tools, refactoring, code completion, type inference.
Description:
During the development of a program in any programming language, the IDE has a very important role. It will never replace the knowledge of the programmer but it can make his life easier. As Smalltalk doesn't have explicit type information, the IDE can not perform the same mechanisms that are present in IDE like Eclipse or Intellij IDEA. These mechanisms can be very useful. The idea of this project is to add type inference mechanisms to try to provide more information to the refactoring and code completion tools. It is important to say that the idea is not to check types or correct the program, but to allow the programmer to have more information.
On Thu, Feb 13, 2014 at 12:06 PM, Usman Bhatti <usman.bhatti@gmail.com>wrote:
[Anne Etien could not post this msg on the mentors list I m forwarding it on her behalf].
Title: FAST JavaScript model
Level: advanced
Possible mentor: Anne Etien
Possible second mentor: Nicolas Anquetil or Yuriy Tymchuk
Description: For in depth source code analysis a support of abstract syntax trees is required. FAST is an abstract syntax tree extension for FAMIX meta-model that is used by Moose technology. The goal of this project is to create a JavaScript version of FAST.
Technical Details: As programming languages are different, their AST representations are different too. FAST model aims for creation of as generic as possible core that can be extended for different languages. Also the structure of a model allows creation of generic algorithms like symbol resolution, metrics calculation and rule checking that will work for any language. Prototypes of FAST for Smalltalk and Java are already implemented. During the project a student will implement the JavaScript model. More detailed information about FAST is provided in the Moose day presentation: http://youtu.be/dRr3WHOD3x4
Benefits to the Student: The student will gain a deep understanding of a JavaScript syntax and abstract syntax tree model. He will also learn about PetitParser framework, and gain knowledge about software modeling and analysis.
Benefits to the Community: Community will get a FAST model for JavaScript that can be used for software assessment with Moose. Also this model can be used later in PhD projects such as automation of source code translation. With this third model, the community will have raw material to real think and improve generic algorithms based on AST.
On Tue, Feb 11, 2014 at 10:42 AM, Damien Cassou < damien.cassou@gmail.com> wrote:
Hi fellow Pharo hackers,
ESUG, the European Smalltalk User Group, is applying for this year's Google Summer of Code. As you probably know, the Summer of Code provides the opportunity to fund students to work during the summer on Pharo. Please reply to this email (be sure to use "Reply to all") if you have ideas you would like to propose.
Please include a summary of the project and links to web pages that can help prospective students to write their application. Please also include the following information:
- if applicable, other dialects that you would be willing to mentor this project for
- the skill level
- name of the mentor(s), email addresses, and possibly any IRC network/channel/nickname where they can be found.
Thanks for contributing to ESUG's Summer of Code application!
-- Damien Cassou http://damiencassou.seasidehosting.st
"Success is the ability to go from one failure to another without losing enthusiasm." Winston Churchill
-- You received this message because you are subscribed to the Google Groups "Smalltalk GSoC mentors" group. To unsubscribe from this group and stop receiving emails from it, send an email to smalltalk-gsoc-mentors+unsubscribe@googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.
-- You received this message because you are subscribed to the Google Groups "Smalltalk GSoC mentors" group. To unsubscribe from this group and stop receiving emails from it, send an email to smalltalk-gsoc-mentors+unsubscribe@googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.
-- Pablo Tesone. tesonep@gmail.com
-- best, Eliot