High performance Git

#675 – May 03, 2026

it is also a content-addressed database, a filesystem cache, a graph walker, and a transfer protocol

High performance Git
a few hours by Ted Nyman

Git is more than version control. It's also a database, a cache, a graph walker, and a transfer protocol. Ted's book covers all those layers and what they cost in performance, from objects and history to packfiles, cloning, and transport. It's aimed at engineers who need Git to stay fast as repos and teams grow.

Shrink Your Backlog: Build Accessibility In, Don’t Bolt It On
sponsored by Level Access

Development leaders agree: Insufficient time to fix issues is the #1 challenge blocking
digital accessibility progress. And the longer issues sit in your backlog, the harder it
becomes to address them. So how can you stop chasing bugs and start preventing
them? Check out our on-demand webinar for tools and tips you can use to clear your
backlog for good.

How the does Shazam work?
8 minutes by Shri Khalpada

Shazam ignores melody and lyrics entirely. Instead, it converts sound into a visual map of frequencies, then keeps only the loudest peaks, like a connect the dots picture of the song. Pairs of those peaks generate short codes that act as a fingerprint. Your phone sends those codes to a server, which finds a match across millions of songs in milliseconds by looking up each code directly rather than scanning every song one by one.

You can beat the binary search
7 minutes by Daniel Lemire

Binary search beats linear search for large sorted arrays, but modern hardware can do even better. By splitting search ranges into quarters instead of halves and using CPU instructions that compare 16 values at once, a new approach called SIMD Quad cuts search times in half or more compared to binary search.

The 20 software engineering laws
25 minutes by Milan Milanović

Software engineering has a set of timeless laws that explain why projects fail, teams slow down, and plans drift. These laws, some over 60 years old, are really about human nature, not technology. Adding people to late projects makes them later, work expands to fill available time, and any metric that becomes a target stops being useful. Learning these laws before a project goes wrong is far cheaper than learning them after.

An AI agent just destroyed our production data
10 minutes by Jer Crane

An AI coding agent deleted a small business's production database and all backups with a single API call, wiping three months of customer data. The agent later admitted in writing that it broke every safety rule it had been given. Two vendors share the blame: Cursor's safety guardrails failed to stop the agent, and Railway had no confirmation step for destructive actions, no scoped API tokens, and stored backups inside the same volume they were meant to protect.

And the most popular article from the last issue was:

newsletters