The compiler platform is currently able to bootstrap on top of an interpreter, and the result of the bootstrap process is able to compile itself.
The backend is capable of generating x86 and MIPS assembly source, a separate tool is used to assemble that source to object code.
The project started out in R5RS Scheme, in large part due to inspiration from Wizard Book, and the author was determined to bootstrap the compiler in that language. Scheme is a useful programming language but has a very small community by modern standards, which might limit the overall interest in the project.
The plan is to adapt more of the components of the project over to a more mainstream language, however keep the Scheme frontend of the compiler written in Scheme. JavaScript seems like an ideal choice given many language similarities with respect to Scheme and the amazing community and tools that have been built up around JavaScript over the years.
The Scheme implementation provided by the compiler and associated runtime library is not trivial, however it does not provide a complete implementation of R5RS. Notably it is lacking a number tower and support for continuations.
Continuation support was planned and there is a continuation-passing style transform module partially implemented in the compiler. However this was not necessary for bootstrap so left for a later project.
Similarly the number tower work was not necessary for bootstrap and left for a later project.
The Scheme frontend is fairly simplistic, and a lot of work needs to be done to make the output of the Scheme frontend more efficient. In particular, eliminating environment allocation would be a big improvement. In general, many of the ideas of the Rabbit scheme compiler were planned to be implemented but are not done for the initial release.
The memory management heap implementation needs some work to allow the heap to grow and shrink. Currently it can be configured at compile time to be large or small depending on the expected working set size.
The garbage collector is conservative and very simple, it uses a mark-and-sweep algorithm. Pointer reversal was planned but not yet implemented. Pointer reversal is needed to make the system reliable, when working on large working sets it is not hard to blow the stack and crash the executables during GC.
The Scheme runtime library needs various improvements, for example it should support some bit twiddling to distinguish integers from pointers at runtime, so that not all integers need to be boxed as objects. Another planned improvement is to represent Scheme objects in the same format as that expected by the Java frontend so some runtime library code may be reused and shared.
The phases of the compiler are all separate tools, currently these phases are not wrapped up into a convenient command-line tool ("driver" is the archaic term among compiler developers for such a tool such as cl.exe). So one has to script the phases of the compiler by hand which is a nuisance. A more convenient tool wrapping the phases was planned but not required for bootstrap.