I feel like I’ve managed to do quite a lot this year and also have a lot planned. With that in mind why not reflect a bit on what was planned, done and what I’m going to be doing.

When relevant links aren’t mentioned within the text body I’ll pop them at the bottom but I might be inconsistent in this. Sorry (not sorry).

rustnation

The start of the year was pretty much dominated by my preparation for Rustnation at the end of March. I’d put forwards an ambitious brief for the talk given the time I had. My talk Creating a Text-To-Speech System in Rust involved a mix of open sourced Emotech code and some code written for the talk.

As well as a fully working repo there was a repo for SSML parsing and a signal-processing based vocoder with small blog/tutorial provided. And a lot of documentation comments about machine learning in general and things I tried that did and didn’t work.

Below is an example doc comment on a neural network model I initially tried to use and abandoned. I kept the source files in the repo and documented them all for anyone really interested. An example of documenting issues with a failed approach:

After getting our ONNX file however it uses the loop operator and has dynamically sized inputs within the model. These are two things that proved fatal to running it in a Rust runtime.

After finding out about ORT I tried it there, unfortunately pytorch can output ONNX that doesn’t obey the standard (-1 dimension size in expand nodes).

There’s an old version of torch used in speedyspeech, one with worse ONNX support. Any changes to the pretrained graph would need to be done via this old version of torch and torch doesn’t come with a clear upgrade path. To run speedyspeech I would have had to port the code to a much newer torch version and retrain a model from scratch.

After all is said and done I think I did a good job and I’ve gotten a lot of positive feedback from people during the conference and even months afterwards. I’ve also seen some of my feedback about DX issues in some ML libraries result in an issue on a relatively new project rten to make sure the next generation of libraries are easier to use (the issue). This to me is a sign that struggling through 4 different ML inference frameworks to find I couldn’t get an open source model working was worth it in the end.

There’s no plans currently to continue this project but I might be tempted to do some future conference work on more statistical less neural-network based models - but no promises!

tarpaulin

Tarpaulin saw 11 releases this year. I have to admit I’ve had periods of burnout working on it so it’s been in a “soft” maintenance mode. I’ve still been fixing issues, updating docs and helping users who post things but I’ve not maintained any meaningful velocity on it and there’s a number of hard issues still to be tackled.

There’s been some improvements with handling toolchains in non-standard places, windows and mac support improved. Logs and error messages also improved to help people debug coverage issues more meaningfully.

One thing I also did was redo the item naming for things like lcov, now they spit out names in the same (albeit assumed unstable) form that cargo-mutants uses. With this change it’s possible with either some manual copy and pasting or script glue to use coverage to exclude untested code from cargo-mutants runs saving testing time. You can see my comment about this in this issue.

Aside from this I’ve mentored one person through fixing an issue they had in tarpaulin. This is unusual but something I’m happy with, a lot of people’s issues are more in the harder problems and the difficulty cliff is intimidating so I don’t often get to mentor anyone like this successfully. It’s always nice when it works out. You can see the PR here.

I did submit a tarpaulin talk to Eurorust but it wasn’t accepted, c’est la vie! After Rustnation I did go for a much less ambitious proposal though so at least I wasn’t too invested!

Next year I have some plans to fix some of the lingering ptrace issues that make tests flaky. Unfortunately, the LLVM coverage is limited in what types of things it can collect coverage for and while it’s generally more accurate if you fall into those niches you’re not going to have results you’re happy with. Additionally things like #[tracing::instrument] attributes cause functions to be removed from the results. Given these issues it may be that ptrace and output formats are the main selling point of tarpaulin against other things like cargo-llvm-cov, gcov or rawdogging rustc llvm instrumentation.

Two years ago I had started to look at using minicov for embedded on-device coverage. I may go back to those experiments, or I think I see a way to use minicov to fix some issues in the llvm profiling runtime when it comes to things like tests with forks/signals/required-panics when a custom test harness is used. I doubt this will be massively relevant to most people though so it may take a backseat - I may just blog about the ideas but I want to test some assumptions first.

dechib

I started work on a RDBMS using rocksdb called dechib. Unfortunately, I haven’t had much time to work on it so I just have table creation, and insertion plus some constraints implemented. Hopefully, next year I’ll have time to implement actual queries - part way through the design for querying from inserted data.

Not much to write about here but hopefully there will be in future! This is just an educational project working on the inner systems that interest me most. Never use it in production if I get it working.

object-trustfall-adapter

Inspired by Predrag I wanted to make a trustfall adapter around ELF files to query ELF files like a database for reverse engineering purposes. Nothing really to share yet, but this is me committing to building in public.

bidirectional audio streaming blog series

In my work we do some audio streaming services for things like live transcription or streaming TTS. However, streaming systems are complicated and it can often be a source of confusion for newcomers to the organisation. Inspired by an actual problem I started a blog series about this which I’ve been posting on the socials™. Four posts are done, one is in late stage draft and another one is in review and another is in early stage draft.

Next year I aim to finish all the planned chapter so publish the 3 in-progress posts and add ones on: observability, testing and the gRPC API. I’m excited for when the metrics post is ready to be published (I am trying to keep to a natural order). I’ve put quite a bit of work into it and it covers a lot that I’ve rarely seen written about.

Longer term (beyond the year), I want to do some tooling around AsyncAPI in Rust that matches some of what we have for OpenAPI/Swagger. Also, a websocket mocking library which I’m starting to draft.

rustybug

I bought an early access copy of Building a Debugger - Sy Brand and have been working through it during my christmas holidays as a means of relaxation! I’m doing it in Rust so it will hopefully feedback back into fixing some of tarpaulin’s ptrace code.

I’m also going to use this to find any opportunities to add PRs to libc or nix if I spot any APIs that need some love. The first of these is already open: libc PR.

Expect some blog posts about this when I have something that works good enough!

outside of technology

Life isn’t just Rust though (or so I hear). Outside of Rust I’ve made decent progress in learning Japanese (for my partners family not weebery… okay maybe also some weebery). I’ll be spending a number of months in Japan as a result and hope to made even greater strides. Also, being a stereotypical software dev I have some fitness goals related to climbing, I got into board climbing last year and will continue with it. I won’t bore you with those details!

I’ve ready a number of Ursula K. Le Guin stories (The Left Hand of Darkness and collected short stories). And the first two Hyperion books by Dan Simmons. I’ll be finishing Hyperion next year, and hopefully also the Wizard of Eathsea series. I’ll also be reading How to Blowup a Pipeline by Andreas Malm, and Abroad in Japan by Chris Broad. This year I’ve kind of read less but the Ursual K. Le Guin short stories book was a chonker.

I’ve watched way too much to enumerate and I’ll probably continue to do that. The next movie I’ll likely watch is Decision to Leave and I’m very excited to see the new Nosferatu.

For gaming I played a lot last year. This year I plan on playing 1000xRESIST, Chasing Static and Hollowbody. I’ll also finish Half-Life which I’m playing all the way through for the first time (I know I know better late than never).

The last year my favourite game I played was SOMA. I might have to play all the amnesia games if they have vibes that impeccable!

Well that’s it for my year in review and what’s coming up. Some of this will be blogged some of it you may never hear about again. Laters 👋