I just published a proposal for a new RSS profile type for Go. It aims to solve the issue that the heap profile typically shows < 50% of process RSS, which is very confusing. It should also help to quickly identify when memory usage is dominated by cgo. github.com/golang/go/is...

We need to talk about excessive nil pointer checks in Go. Writing code that's safe and prevents panics is excellent, but when we reach the point where no pointer can be trusted, we've dug ourselves into a hole.

It's only a discussion right now but opaque + sum types = efficient oneofs github.com/golang/go/is...

Interesting finding from benchmarks, protobuf's Go opaque API was a win for elementary fields, but the gains fell off once those fields were wrapped in oneof.

What's new in gopls 22: go.dev/gopls/release/v0.22.0 Please try it out: $ go install golang.org/x/tools/gopls@latest

Ideally, middleware execution order shouldn't matter, but consider caching, error wrapping, and concurrency limiting. go-redis changed: 2020 FIFO to LIFO: github.com/redis/go-red... 2023 LIFO to FIFO: github.com/redis/go-red... The lesson: in lack of order-invariance, have a test to check for it.

Quality-of-life improvement: adding a test helper package that lets you construct context variants in one line, without anonymous functions in table-driven tests: ctxtest.New(t, ctxtest.WithCanceled(true)) The kind of package you create after wishing it existed for the fifth time.

Remember when it was called source code, not just code?

I don't always make open source contributions, but when I do, it feels like this.

One of my favorite Go proverbs: write code, don't design types.

I find it increasingly judgmental to state that something is an anti-pattern without additional context. It moves the conversation to: you're obviously doing it wrong. I think we can do a lot better.

Before concluding that the fix for issues seen with agentic programming is design by contract, or focusing on writing clear specs that can be thrown over the wall, it is worth revisiting the history of the waterfall model.

How many files is too many for one Go package?

The longer you maintain a large Go codebase with lots of contributors, the more appealing gofumpt gets. Fewer choices, fewer style nit comments, less time wasted.

Just released a new version of my PID controller library in Go. When used well, it can help backend systems adapt smoothly to changing conditions. This release adds metrics and options for easier inspection, tuning, and stabilization github.com/konradreiche...

"This interruption of consciousness is costly... the most serious effect may well be the decay of grasp of all that is going on in a complex system." What happens to that grasp in agentic programming, when we're waiting on agents, or coordinating many in parallel?

I used to think maximum function length in Go just depended on context. Now, if it doesn't fit on the screen, it's too long. Before anyone objects, buying a larger monitor would not count as refactoring 🖥️

"The hardest part of the software task is arriving at a complete and consistent specification, and much of the essence of building a program is in fact the debugging of the specification."

I reread Brooks' No Silver Bullet and found it incredibly refreshing how well its arguments, analogies, and walkthroughs of different technologies still map onto today, 40 years later.

When the internet felt like the wild west: infinite popup windows from a browser exploit crashing your PC, starting a JDK download and knowing it would take 4 hours, getting hit with email bombs, and teaching yourself just enough Visual Basic Script and PHP to feel like you were running the machine.

with enough tokens we can completely convert memory unsafety into complexity and then everything will be simpler because there will only be one kind of bug cmon lets do it why are you looking at me like that dont you like security

Highly recommend this podcast with Charlie Warzel and Jonathan and Melissa Nightingale on AI in the workplace: insights on how poorly curated AI output erodes trust, what humans value at work, and why management is a skill and not just delegation. More in replies: www.youtube.com/watch?v=ncmu...

NEW POST Modern hardware is fast, but software often fails to leverage it. @withcaer.com guides their work with mechanical sympathy. They distill this into principles: predictable memory access, awareness of cache lines, single-writer, natural batching martinfowler.com/articles/mec...

I first ran into PID controllers while working on cache eviction. They're a simple way to keep systems on target as load shifts. I wanted to better understand them, so I reimplemented one in Go and put together a walkthrough: konradreiche.com/blog/an-intr...

Leading Go code review nit: putting helper types or functions above the main exported identifier. It reads like Go needs forward-declaration. Because you can define in any order, let's order by importance. When I open a file I want to see the main characters, not the background actors 🎬

AI-assisted code reviews can be incredibly powerful. But I'm starting to get wary of them as a first pass. If AI decides what you see and question, it'll interfere with your independent thinking and, worse, give a false sense that you've done some of the work, so your own review ends up weaker.

Partial blame to Microsoft and Zoom for normalizing the need to install new software just to make video calls, despite browsers being perfectly capable. They make it extra hard to even find the web-only links. simonwillison.net/2026/Apr/3/s...

"Software is free as in puppies. It will pee in your bedroom and eat your furniture. The weight of every line of code is real. We will need to maintain it. We will need to port it. It goes into the context window. Someone will get paged at 2 a.m. unexpectedly." ― @industrialempathy.com

"In a thoughtful piece, Azeem Azhar, the technology writer, describes his efforts to safeguard 'the space where ideas arrive before they’re shaped'. But how many of us will put in such careful, reflective effort to protect our most generative spaces of thought?"

I was happy to see JetBrains release modern Go guidelines for AI agents. That said, we added go fix ./... to our formatter and CI, and it quietly solved most of it. The Go team keeps surprising by turning rules into deterministic tooling.

Great read from @apenwarr.ca. Reviews don't scale. Each approval step turns minutes into days of waiting. They catch issues, but after the code is written. The system already allowed them. Repeated comments point to missing boundaries and unclear APIs apenwarr.ca/log/20260316

If adding less code feels impossible, that's a signal that we should consolidate what already exists instead of extending it. AI increases how fast we can add solutions. It does not increase how much system complexity we can hold in our heads.

Watching capable engineers ship large amounts of AI-generated code at an impressive velocity. It was never hard to add more code. The hard part is adding less. We optimize locally. A new problem appears, so we add another piece of code that solves it. The system becomes harder to reason about.

„There are two ways of constructing a software design: One way is to make it so simple that there are obviously no deficiencies, and the other way is to make it so complicated that there are no obvious deficiencies. The first method is far more difficult.“ — C. A. R. Hoare

I've just released a new version of the "sh" #golang module, including initial Zsh support in shfmt, as well as many improvements and fixes to the parser and interpreter 🎉

📣 The #GopherCon Call for Speakers is officially closed! A huge thank you to everyone who submitted a talk proposal. We received 150+ submissions, and our Evaluation Committee is now diving into the review process. Stay tuned as programming is expected to go live in early May. #RoadToGopherCon

This one is pretty funny go.dev/issue/77934. Turns out a struct used for Go ←→ Linux syscalls use int64 when Linux expects int32. But by chance this can't cause a bug because: 1. it is the last field of the struct 2. the size of the struct doesn't change since C aligns int64 to 8 bytes on 32 bits

Less than 24 hours until the @gophercon.com call for speakers closes. I started editing my proposal, then remembered the German word "verschlimmbessern" (to improve something for the worse). Turns out it was already where it needed to be. Threw out the edits 👋🗑️

Excellent example of how a data race can produce a value that was never assigned. This is what undefined behavior can look like in practice: go.dev/play/p/HuIoL...

Very cool Go library from @dronnix.bsky.social. It implements a Black-White Array, an ordered structure with O(log n) operations but far fewer allocations than B-Trees. Cache-friendly and supports duplicates github.com/dronnix/bwarr

🎤 The #GopherCon stage is calling… Dreaming of your mic drop moment? Just 2 weeks left to wow us with your #golang talk proposal! Submit by March 4th 👇 sessionize.com/gophercon-20... #RoadToGopherCon

“Using go fix to modernize Go code” by Alan Donovan — https://go.dev/blog/gofix #golang

It's conference talk proposal season again! In 2024 I had the privilege of reviewing 233 proposals as part of the GopherCon review committee. Most scored below 3/5, not because they were bad, but because they missed the critical piece. I wrote up what I learned: konradreiche.com/blog/what-go...

A colleague shared this Go library that can redefine functions at runtime. I assumed it was a joke. It was not a joke. I'm going to pretend I never saw this github.com/pboyd/redefine

I'm putting the finishing touches on the Staticcheck update for #golang 1.26. You can expect a release by the end of the week.

#GopherConSG is back for 2026! Our next conference will take place on 20–22 May, and our #CFP is open right now! 💥 If you have #golang ideas to share, we're looking forward to hearing all about them! 🙋‍♀️🙋‍♂️ www.papercall.io/gopherconsg-...

Looks like @gophercon.sg is scheduled for May 20–22 🇸🇬 2026.gophercon.sg

Interesting expansion of generics in #golang 👀

I'd been feeling pretty good about leveling up my Vim refactoring skills with regex find/replace and :argdo. Then I wrote my first gopatch, applied it across the codebase, and there's no going back github.com/uber-go/gopa...

Sitting at home in San Francisco, drinking coffee from a Stuttgart Gophers mug that was given to me at the @golab.io conference in Florence, Italy. I find this quite telling of how international the Go community is.