GSOC 2017 Ideas

From D Wiki
Revision as of 04:24, 9 February 2017 by CraigDillabaugh (talk | contribs) (Its Good To Know)
Jump to: navigation, search

This is the D Google Summer of Code page for 2017 - it is currently under construction. If you are interested in participating in the 2017 GSOC as either a student or mentor, and want to do something related to D, please feel free to contact our GSOC administrator Craig Dillabaugh (firstname dot lastname at gmail dot com).


Timeline

The timeline for GSoC 2017 can be found here.

Ideas

Plenty of challenging and important projects exist in the D world. They range from writing new or improving existing modules of D's standard library (Phobos), working on its compilers (Compilers), shaping GUI libraries for D, integrating D with other languages and more.

Mir Project


The Mir project is developing numerical libraries for the upcoming numeric packages of the Dlang standard library. The are numerous projects an interested student could pursue:

  • University project. Your GSoC project can be combined with your scientific research.
  • ndslice>-<[http://julialang.org Julia) integration
  • ndslice>-<[http://www.numpy.org NumPy) integration
  • General purpose BetterC libraries
    • I/O betterC Dlang library
    • Async betterC Dlang library
    • String and format betterC library
  • mir-random
    • Port mir-random uniform floating point implementation to clang and g++. You will code C++, our longterm goal is C++ standard.
  • mir-cpuid
    • ARM support
    • Advanced OS specific support
  • mir-glas
    • BLAS Level 1 subprograms
    • BLAS Level 2 subprograms
    • One BLAS Level 3 subprogram. It is OK to do only one! But it must be faster then OpenBLAS, and not slower then Intel MKL. We know how to do it.
    • Multithreading support for BLAS Level 2
    • Multithreading support for BLAS Level 3

You can get more details on projects related to Mir here.

Its Good To Know

To work on the Mir you should be proficient with one of:

  • C
  • C++
  • LLVM
  • Fortran
  • Experience with D is not essential.

To work on the Mir project requires a responsible and self-motivated student.


Proposed Project Mentors: Ilya Yaroshenko Sebastian Wilzbach Joseph Rushton Wakeling (mir.random)

SDC Project - D Compiler as a Library


Project Description

The SDC project (https://github.com/deadalnix/SDC) is an effort to provide a D compiler as a library. Any ideas to further the development of this project are welcome, but for a student who would like a specific project we propose the following

  • Start by implementing with @property feature of D. This feature will allow a D programmer to create functions that are called using the same syntax as variable access.
  • Using the @property feature the student will be able to implement the runtime support for slices and associative arrays. The operations to implement are as follows:
    • Implement arrray operations like concatenation and appending, and implement a sound memory management strategy for the underlying data.
    • Implement a generic and efficient hash table. The data structure and algorithms used must be flexibile enough to be adapted to any type of data that might be stored in the table. A concurrent version of the table is need for shared data.
  • Finally, the student will implement masquerading of D syntax into calls for the runtime.
  • Integrate LLVM's new JIT infrastructure in SDC, the On-Request Compilation JIT (ORCJit) API. This would simplify the implementation of certain D features such as Compile Time Function Evaluation (CTFE) for SDC.

Its Good To Know

  • Please watch Amaury's DConf talk on SDC.
  • SDC is developed in D (of course) so you will need to be proficient in D by the time you start coding.
  • You should have taken at least one course on compilers, or at the least be willing to educate yourself in this regard. There is a decent course availabe through Coursera https://www.coursera.org/course/compilers
  • You should familiarize yourself with classicial data structures for arrays and have knowledge of various schemes for table implementations, (it is worthwhile to read up on hopscotch and robin hood hashing).
  • SDC uses LLVM for code generation, so some familiarity with LLVM will be required (see http://llvm.org/docs/tutorial/index.html).

Proposed Project Mentor: Amaury Sechet

FlatBuffers Support and/or Improved Protocol Buffer Support


Project Description

FlatBuffers is an efficient cross platform serialization library for C++, C#, C, Go, Java, JavaScript, PHP, and Python. It was originally created at Google for game development and other performance-critical applications.

Currently there is no support for D, so this project would involve building FlatBuffers support from scratch. The goal of the project is to contribute the D support to the upstream repository.

Regarding Protocol Buffers, existing work has been done to provide support for D, however, there are a number of areas that can be improved including:

  • comments and deprecated fields handling
  • generate interfaces for protobuf services
  • upstream repo CI integration
  • define the native D representation of all "well known" types
  • optimization of non-recursive messages
  • benchmarking

Its Good To Know

It would be worthwhile to have experience working with either Protocol Buffers or FlatBuffers in another language, but such experience is hardly a necessity.

Proposed Project Mentors: Dragos Carp

Lowerer


Project Description

This tool would "translate" compilable D code to compilable D code. See initial idea discussed in Issue 5051. It would offer a variety of lowering services for purposes of tooling, debugging, and project management:

  • do not output function bodies, .di style
  • write deduced attributes for functions (useful when function bodies are not written)
  • expand all possible mixins in the code
  • execute lookup on all symbols and write the full symbol, e.g. writeln becomes .std.stdio.writeln
  • explicitate all user-defined operators, e.g. a + b becomes a.opBinary!"+"(b)
  • replace all wholesale imports with detailed imports that specify the symbols needed
  • write specialized versions for all templates used within the module (this is likely to be tricky)
  • lower all scope statements into try statements
  • lower all foreach statements into for statements
  • specify the exact symbols needed for each import statement (which means: if no symbols, the import is redundant - easy to mark as a warning by a subsequent tool)
  • evaluate all static ifs possible
  • lower code using version(), i.e. make the unused branch disappear
  • make all comments disappear
  • make only non-documentation comments disappear
  • evaluate all possible CTFEs (tricky)
  • introduce named values instead of temporaries wherever order of evaluations is defined.

Proposed Project Mentors: TBD

fork()-based Garbage Collector

Project Description

Gustavo Rodriguez-Rivera and Vincent Russo authored a paper on an interesting take on a garbage collector: leverage efficient fork() implementations, which in many Unix systems take advantage of hardware-provided copy-on-write semantics for duplicated memory pages across processes.

This idea is already leveraged in the high-performance garbage collector for D implemented and used by Sociomantic. (A lingering issue is fork() and malloc() share a lock, which is problematic.) Leandro Lucarella, the engineer who wrote the implementation, has open sourced it here, but that is a bitrotten version that has fallen on the wayside.

Leandro would be glad to assist with questions by a motivated implementer. Gustavo has quite a few ideas for improvements, including a possible Windows implementation, and may be able to even coauthor a paper.

Its Good To Know

  • TODO

Proposed Project Mentors: TBD

Linux debugger

Project Description

ZeroBUGS is a high-quality, source-level debugger for Linux implemented from first principles in C++ by Cristian Vlasceanu. The author got busy with work and the project has since bitrotten, as did a fork of it by a different engineer.

ZeroBUGS presents amazing opportunities for D/Linux debugging, and Cristian is willing to guide a motivated implementer.

Who's (Using) Who?

It happens often that executables include code that seems unused (e.g. a typical "hello world" links in functions that are not easily explained). A tool that shows dependency chains would be a great helper in understanding what dependencies are at work, and would give insight into how to reduce them.

The tool would output for each function all symbols it uses. The tool's output would be in one (or more) popular format of stock tools for graph drawing, such as DOT, Graphviz, Sage, PGF/TikZ, newGRAPH, etc.

This can be done using Valgrind's plugin, Callgrind, as explained here.


Proposed Project Mentors: TBD

Phobos: D Standard Library


There are tonnes of opportunities to contribute to improving D's standard library, Phobos. Here we list some modules that might make good projects:

std.benchmark

Ideally every function in Phobos should have its own benchmark suite which is run for every PR to check for performance improvements or regressions.

std.i18n

Design and implement a basic internationalization framework. It may be possible to implement this with pragma(msg). For proof of concept see http://arsdnet.net/dcode/i18n.d . It should provide at least the following functionality:

  • A locale part, std.i18n.locale which should detect a user's default lanaguage, select a global application lanaguage, and provide types to describe and work with locales.
  • A text translation part, std.i18n.translation, which should be gettext compatible, preferably using the gettext serialized hashtables in .mo files, and provide low level (gettext like) and high level (boost::locale like) APIs.
  • A tool to extract strings which need to be translated from D code. This should preferably be based on DScanner but alternately could use regular expressions. Optionally support for D code could be added to xgettext.

std.parallelism

std.parallelism needs a review and some benchmarking - prior to making improvements. As part of this is would be good to have a standard benchmarking framework, hence the idea of std.benchmark. However there is no need for it to be in std (and hence Phobos) in the first instance. So the project(s) would be to create a comparative benchmarking framework that can then be used to analyse std.parallelism on a more scientific basis than has been done to date.

std.units

Unit conversion library

std.serialization

A flexible (de)serialization framework that can be used as a standardized building block for various serialization related things: std.json, std.csv, Protocol Buffers, Cap'n Proto, vibe.d... One important goal would be to define how a user-specified type has to work in order to make it serializable, as well as to allow the end-user (e.g. a user of std.json) to change the serialization of third-party types that cannot themselves be modified. A good starting point is would be to work with the Orange framework.

std.container

  • support for custom allocator (aka no GC)
  • thread-safe
  • performance
  • existing work: here and here.
  • Additionally lock-free data structures (e.g. based on Martin Nowak's work would be a huge benefit.

std.database

  • Provide a general database interface for D
  • Replace etc.c.sqlite with a native implementation
  • Existing work: here, and here
  • See also the following DConf talk and this one and this one.

std.decimal

A general decimal math library for fixed point arithmetic.

std.encoding

std.encoding needs a complete overhaul. Other encoding packages as ascii, base64, utf8 could be included and a common API should unite them.

std.eventloop

High-performance native event loop abstraction for D.

std.graph

Graphs are used in all sorts of contexts - from low-level network communication to high-level data science.

  • At the moment there is only one unmaintained D library, but

Boost::graph would be a good direction.

Ideas From Previous Years

GSoC idea pages from past years:

Tips for students

Daniel Pocock has written a detailed blog about initiatives students can take if they want to have a serious chance of being selected in GSoC without a focus on one specific organization.

To learn more about potential mentors check our mentor page here.

Tips for Mentors

If you are interested in mentoring, please check out the organization administrator and mentor manual for more information.