SharpLang

SharpLang is a .NET (MSIL) compiler front-end for LLVM.

It allows you to compile C# and other .NET languages code to any supported LLVM backend.

Licence

BSD 2-clause license.

Documentation

Getting Started

Objectives

• Fill the gap between managed flexibility/safety/productivity (on by default) and native speed/control (when you actually need it)
• Supports many platforms (including Windows, MacOSX, mobile devices, consoles, browser)
• Performance oriented
• Lightweight runtime
• Some .NET features that implies too much overhead for to their actual uses will be emulated or dropped (i.e. array covariance -- unsafe and huge performance implication)
• Various technical choices to avoid unecessary conversions and interop with other languages
• Provide a good starting point to implement "M#" features as soon as possible (http://joeduffyblog.com/2013/12/27/csharp-for-systems-programming/)
• Flexible and controllable (.NET shouldn't be a black box anymore; optimize what you want, pay for what you actually use, etc...)

Planned features

It's still a early work in progress. Any help/feedback is welcome!

Short term

• Supports for all .NET opcodes
• Passes many unit tests (branching, struct/class, interfaces, generics, etc...)
• Subset of .NET framework (maybe .NETCore 4.5 to start with?)

Future additions

• Immix GC (TBD)
• Performance
• Non-nullable types
• SIMD support
• More control over class/struct layouting (allocate class on stack, etc...)
• Various "M#" improvements
• Uniqueness and Reference Immutability for Safe Parallelism: http://research.microsoft.com/pubs/170528/msr-tr-2012-79.pdf
• Lifetime control (unique/owned pointers, stack alloc as much as possible with pure functions, etc...)
• RAII (struct destructors)
• Use and extend Roslyn for new language extensions
• Direct (or at least easy) access to C++, probably with help of libclang

Various technical details

Here for now, until later moved to a more specific documentation.

• UTF8 string (UTF16 was choosed mainly because of Windows API, but UTF8 is more space efficient, API compatible with char* and avoid many conversions)
• String/Array indrection to more easily carve substring/subarrays.