Tweag is delighted to introduce our first cohort of Tweag Open Source Fellows.

The Tweag Open Source Fellowship is an opportunity for open-source contributors of all stripes and backgrounds to be paid for their hard work. Fellows are selected twice a year (applications due March 30 and September 30), and then they work with a mentor for 12 weeks of the Fellowship. Mentors are full-time Tweag employees with an interest in the open-source project being proposed.

We received 19 applications by our application due date of March 30. These applications went through a three-stage reviewing process. First, each application was reviewed by at least two members of the Fellowships team, to check that the project was feasible to fund. At this stage, we looked for projects with a reasonable chance for success in 12 weeks and that were well-defined and impactful enough to consider funding. Twelve applications went to the next round.

These remaining applications were each assigned to at least two full-time Tweag employees not on the Fellowships team, for in-depth reviews. Using these reviews, the Fellowships team selected seven applications for our shortlist. Key factors at this stage were our ability to identify knowledgeable Tweag-internal mentors and how well projects aligned with broader Tweag interests. The shortlist was reviewed with Tweag’s admin team, resulting in four offers, sent out on 5 May, 2020. All four offers were accepted, and we’re delighted to welcome four Fellows to work with us:

Avi Dessauer

For his Fellowship, Avi will be implementing Sundial GC, his design for a pauseless, concurrent, copying garbage collector. Sundial is designed around a novel approach that exploits monomorphism and immutability in order to offload GC work onto dedicated threads, while minimizing inter-thread synchronization. Sundial thereby eliminates the bane of low latency systems: GC pauses. Sundial’s immediate goal is to enable functional programming in safe Rust, with the eventual aim of being embedded in future low latency functional language runtimes.

Avi is a self-taught polyglot systems developer from upstate NY. He contributes to numerous open source projects, primarily in Haskell and Rust. He fell in love with FP, and Haskell specifically, a few years ago. Unfortunately, Avi’s other passion, distributed database design, is latency sensitive. Therefore, he set out to solve GC latency in copying collectors. In addition to CS, his interests include political science and cooking.

Avi will be mentored by Tweager Eelco Dolstra.

Zubin Duggal

The goal for this Fellowship is to build a responsive and featureful Haskell editor experience by contributing to successful efforts on the Haskell Language Server. Zubin plans to do this by taking advantage of the indexing and caching capabilities offered by HIE files and hiedb. These will support fast, project-wide queries, such as reference lookup, jump-to-definition for dependencies, and best-effort IDE support even in the face of compile failures, so that your IDE can aid you when it is most needed, such as when you are in the middle of a big refactor. Zubin also hopes to make ghcide much more scalable and responsive with these changes.

A part of this work is to help in the transition from haskell-ide-engine to haskell-language-server, so that the Haskell community can build on ghcide to offer a featureful IDE, as well as a platform for tooling developers to easily integrate their tools into and get them in front of users with minimal effort.

You can follow progress on this project with weekly updates at https://mpickering.github.io/ide/

Zubin is currently a Master’s student in Computer Science. He has been involved with Haskell IDEs since 2017 when he completed a Haskell Summer of Code project which led to the first release of haskell-ide-engine in its LSP incarnation. Since then, Zubin has completed two more GSoC projects, allowing GHC to emit .hie files and working on hiedb, haskell-ide-engine, hie-bios, cabal-helper and other tools in the ecosystem.

Zubin will be mentored by Tweager Cheng Shao.

Guillem Marpons

This project is aimed at enhancing Pandoc with reading support for the markup language Asciidoc, and also improving Pandoc’s Asciidoc generation. The parser Guillem proposes to develop is not a regular Pandoc Reader, but something more similar to commonmark-hs: a Pandoc front-end whose syntax can be easily extended, tracks accurate information about source positions and can potentially support all Asciidoc features.

Guillem is a software developer with more than 20 years of experience and also some background in research. He has always been a FLOSS advocate and has devoted the last few years to providing guidance and training on FLOSS adoption for public institutions and projects. He has also implemented documentation strategies based on lightweight markup languages and docs-as-code principles in those projects.

Guillem will be mentored by Tweager Juan Raphael Diaz Simões.

Georg Rudoy

Dependent types are a powerful tool allowing expressing almost arbitrary specifications about a program’s behavior and proving that the program, in fact, follows its specification. Unfortunately, using dependent types also requires the programmer to explicitly write down the proofs involved, which is certainly hindering their wider adoption. On the other hand, refinement types are a subset of dependent types enabling decidable and efficient proof automation, and, although their expressive power is nowhere near that of full dependent types, they still are of tremendous help for day-to-day programming. The natural question then is whether it is possible to seamlessly unify both in a single language, compiling refinement types to the base dependently typed language. The aim of this project is to investigate whether this idea is viable by means of a proof-of-concept toy language implementing a flavor of refinement types, with a translator to Idris. This language and the translator will then serve as a basis for a later development introducing refinement types into an existing dependently typed language.

Georg is a software engineer whose projects over the last decade ranged from machine learning to compilers for domain-specific languages. Lately, his interests shifted towards provably correct programming and branches of mathematics like type theory. His production development experience together with his eagerness to do more research-oriented work means he is excited to begin this project.

Georg will be mentored by yours truly, Tweager Richard Eisenberg.


Interested in applying for a Tweag Fellowship? Our next deadline will be on September 30, but the application is open at all times.