https://github.com/thaliaarchi/nebula2
Toolchain for the Whitespace programming language
https://github.com/thaliaarchi/nebula2
compiler whitespace-language
Last synced: 2 months ago
JSON representation
Toolchain for the Whitespace programming language
- Host: GitHub
- URL: https://github.com/thaliaarchi/nebula2
- Owner: thaliaarchi
- License: lgpl-3.0
- Created: 2022-07-05T05:19:47.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2023-06-09T05:08:38.000Z (almost 2 years ago)
- Last Synced: 2025-01-25T10:08:33.687Z (4 months ago)
- Topics: compiler, whitespace-language
- Language: Rust
- Homepage:
- Size: 260 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Nebula 2
Nebula 2 is a library-oriented compiler with a dynamically-registered
mixed-dialect IR.
The project is very early in development, but the aspirational architecture plan
is as follows:
## Aspirational architecture
A language can be defined and dynamically registered with its own syntax(es),
including accurate location tracking in all phases; intermediate language
dialect(s); and incremental transformation passes between arbitrary dialects.
Most languages will be able to serve as a frontend or a backend, using
incremental transformations that lower to or raise from common dialects.
Dialects decouple control flow, data flow, and memory models, to support diverse
representations. The core data flow dialects will include an SSA-based IR, that
uses basic block parameters, and a typed stack-oriented IR. Control flow
dialects will include unstructured with jumps and structured with functions and
loops. There will also be eager values and lazy thunk expressions, as well as
arbitrary-precision integer types, with a data flow analysis to constrain bit
widths.
The dynamic registration allows languages to be defined modularly in independent
projects and for library components to be reused by multiple tools, including
external tools. The toolchain will include a static compiler and a language
server.
I intend to use incremental computation via [Salsa](https://github.com/salsa-rs/salsa),
as [used](https://rustc-dev-guide.rust-lang.org/salsa.html) by rust-analyzer, or
[Adapton](https://github.com/Adapton/adapton.rust) ([notes](https://github.com/thaliaarchi/notes/tree/main/topics/adapton)),
once I have a complete pipeline.
It takes inspiration from many projects and concepts: MLIR for dialect mixing
and registration; GraalVM for its polyglot API with high-level mixed-language
interoperability and its sea of nodes IR with loose node ordering; LLVM for its
library design and SSA-based IR; the SSA form variant in used in MLIR and
WebAssembly, that uses basic block parameters instead of phi functions; and
Futamura projections for powerful partial evaluation.
As I've been thinking about the kinds of programs I want to write, I've been
formulating ideas for languages that I would design in tandem. One would be
concatenate with quotations (closures) like Factor and a statically-typed stack
like Kitten.
## Philosophy
I've been developing it forwards with small instruction sets, similar to the
Nanopass philosophy, except I am compiling complete small languages that have
shared features, rather than building up subsets of one language. I've started
with Whitespace, of course, which I consider to be a nice starting point, since
its jump-oriented control flow resembles a minimal LLVM IR, which I can expand
into a general IR with features added as needed, until it approaches parity with
LLVM IR. By writing transformations between structured and unstructured control
flow, I could eventually use LLVM IR as both a back end and a front end.
## Other projects
Nebula 2 is the successor to [Nebula](https://github.com/thaliaarchi/nebula),
an optimizing Whitespace compiler, written in Go, that uses static
single-assignment form for its own intermediate representation and lowers to
LLVM IR.
The [Whitespace Corpus](https://github.com/wspace/corpus) is a collection and
comparison of other projects for Whitespace and aims to catalog every
implementation.
## License
Nebula 2 is released by Thalia Archibald under the terms of the
[GNU Lesser General Public License](https://www.gnu.org/licenses/lgpl-3.0.html),
version 3 or later. This license makes the it free to use, share, and improve,
with the requirement that any alterations or improvements to Nebula 2 also must
be released as open source under this same license.