February 2022 readings

A bit more than a month and a half elapsed since last post. I haven’t been very diligent in writing here, so there are a bunch of things I read and watched that didn’t make it in this post.

General computer things

• Uninitialized Memory: Unsafe Rust is Too Hard, by Armin Ronacher.

  This is a feeling I share. Zig is proposing some genuinely new ideas there. Alignment is easier to work with (@divExact). Niches, etc. There is a lot to say about finding the right amount of static analysis, and making dynamic checks tunable. Nothing replaces knowing what your code does in practice. Knowing what is in memory, and where is also non-negotiable.

• Read the JSON-RPC 2.0 spec end-to-end. It’s surprisingly small. I like this.

• Out of the Tar Pit, by Moseley and Marks, 2006.

  • Complexity is defined as what prevents understanding the system. Our understanding of our systems has two major sources: testing (in the broadest sense) and informal reasoning. Informal reasoning is always there, always used. It’s most prudent to use both testing and reasoning.

  • One large source of complexity is mutable state: “have you tried turning it off and on again?”

  • Control flow between statements vs. relationships between data.

  • In most contemporary systems, complexity increases more than linearly with size.

  • Simplicity is hard… Power corrupts. The authors argue that increased power, like manual memory management, will inevitably lead to abuses of that power. I disagree that the power sits at the level of what the language allows, it’s what the system allows — if you take into account managing laziness and knowing your runtime in Haskell for example, you end up with something more complex in that regard, than an equivalent imperative program without garbage collection.

  • OOP has difficulty of representing multi-object constraints. Quote from Baker 93:

    In a sense, object identity can be considered to be a rejection of the “relational algebra” view of the world in which two objects can only be distinguished through differing attributes.

  • The paper then loses itself in comparisons of ideal vs real world, accidental and essential complexity.

  • It then comes to the interesting part: the relevance of the relational model outside of databases. Data independence, relations, access path independence, etc.

  • The last part of the paper presents how the authors propose to combine the functional and relational models — FRP (functional relational programming).

• The Data Oriented Design book.

  Less stuff in memory, neatly organized in arrays, no pointers.

  The chapter on optimization is a really well-constructed, well-argued framework on when and how to optimize.

  The book is worth a read for at least three reasons:

  • It makes the connection between data-oriented design and techniques from the database world very clear.
  • It is chock full with small bits of wisdom and techniques you might not have thought of.
  • It structures practices and information many of us knew, but hadn’t organized quite as well.
  • Because of that last point, it is also a good introductory resource to data oriented design.

• Handles are the better pointers

  General idea that is well known to Rust programmers. Twists, and ideas for encoding more in the handles.

Zig

Two talks:

• The road to Zig 1.0 by Andrew Kelley — would recommend as an entry point to Zig, the why and the how.
• What’s a Memory Allocator Anyway? by Benjamin Feng. Good high level refresher, would recommend if you want to introduce someone to the topic.

As mentioned higher up on this page, I think Zig has a bunch of genuinely good ideas (tweakable runtime checks, build ergonomics, etc.). If it pans out, it is going to be a large scale experiment on staged compilation — is it a viable medium for reuse and abstraction, as an alternative to more advanced type systems?

Gemini

• Gemini Is A Little Gem, by Andre Garzia.
• The Gemini spec and related documents.

Impressions: need to give it more of a try, but seems limiting (no math?). A bit too much novelty for novelty’s sake. The web may be broken, but at least it’s backward compatible. Blogs like this one can still exist.

Databases

• FOSDEM talk by Shlomi Noach on new schema migration features in a Vitess context

  Video

  After minute 10, there’s a good summary of “why down migrations are not a great idea” (matches our reasoning).

  This is good background on what PlanetScale migrations are built on.

• Databases: How large is too large? by Kristian Köhntopp.

  Just delightful short post on good practical limits for database instance sizes given current hardware.