I kinda had to learn Rust, because I sucked at C/C++ so badly. Once beyond the fun phase of copy-and-paste Arduino programming, my fascination with microcontrollers inevitably threw me back into the C/C++ nightmare of my college years. I could do Java/Scala/Python, but still clueless about how to properly manipulate strings in C/C++. So Rust was kinda my lifeboat.
And in order to learn Rust, I had to do a real project. After a few false starts, I stumbled upon a neat little project called xsv that can JOIN csv files with blazing speed. Kinda like what we use Spark for at work, except condensed into one tiny CLI tool. Eventually I found the one active fork called qsv maintained by friendly Joel and decided to jump in.
Here are a few takeaways from my very first real Rust project.
No need for Rust Debugger
I sunk a few days into trying to get the LLDB and GDB debugging flow working. The payoff was too small. I am sure I would need such powerful tools someday, but for my simple and occasional needs, dbg!() is enough.
rustfmt is a Friend, and clippy is a Mentor
I received some complaints on my first few PRs because I wasn’t used to running cargo fmt. Same with cargo clippy. Then I found that I could be sloppy with my whitespace and indentation as long as I run rustfmt later. And clippy made me feel like having a personal mentor standing over my shoulders to check my newbie Rust code.
Cargo Check on Save
I had disabled cargo-check-on-save for a while because my laptop would almost freeze with all the work it was doing. Then I read the first section of Zero to Production in Rust and decided to give it another try in order to decrease perceived compilation time. It made a night-and-day difference to my productivity! Compilation errors appear immediately after making changes, all visually annotated right with the source in VS Code.
It’s Okay to Upgrade the Compiler
I don’t ever remember upgrading the compiler so frequently in Java or Python land. Just a few days after a new Rust release, qsv would actually pick up new Rust language features, and I would be forced to upgrade my Rust version in order to compile. With Rustup, it just worked.
Rust likes Memory
It’s easy to go over the default stack size of 2MB with large arrays [u8] instead of vectors Vec<u8>.
And because Rust works so well, sometimes we forget and try to load really large files that exhaust hardware RAM. It’s amazing that there’s no segmentation fault!
Serde is Slow, and so is Clone()
Serde is used pervasively in Rust. Convenience is costly though. Fast code needs to avoid serde.
Sprinkling clone() around is probably the easiest way to silence borrow-checker errors. But payment may come due in terms of performance. Write correct Rust code without resorting to clone().
Rayon and Streaming Data
It would seem that Rayon parallel processing wouldn’t work with streaming data because data would be arriving one at a time. I learned from More Stina Blog! that streaming data can first be put into batches, then Rayon parallel iterators can plow through them as usual.