Old hacker. An eventually inconsistent, leaderless RAFT. Mostly available, occasionally partitioned.
You can find me at Mastodon, Bluesky, LinkedIn, and GitHub.
Québec, Canada.
Roberto Selbach
Blog posts
- — 谢谢你，中国！ (Thank you, China!)
- — Updated LLM Coding Workflow
- — Your App Subscription Is Now My Weekend Project
- — Being Specific when Pairing with Bots
- — Thoughts on Amp's ad-supported business model
- — LLMs Are Tools, Not Replacements
- — Trying out Codex CLI
- — Cedilla in Fedora 2025
- — How to Make :x Just Save Instead of Save and Quit
- — X-Touch Mini for Flight Simulation
- — Impressions on the Keychron Q6
- — In which I reminisce about the last few years
- — The Last of Us Part II
- — Try was worth trying
- — The argument over try
- — I think I like try
- — Zero values in Go and Lazy Initialization
- — Playing with Go module proxies
- — Introduction to Go Modules
- — New licence plate
- — Playing With Go Modules
- — An LRU in Go (Part 2)
- — An LRU in Go (Part 1)
- — How to use FileServer with Gorilla’s Subrouter
- — SciTech links (November 4 2017)
- — Science and tech links (October 21, 2017)
- — Caddy and the Importance of Licenses
- — Returns in Go and C#
- — Happy Birthday Canada
- — SciTech Links (June 5, 2017)
- — Casting objects and boxing
- — Retreating back into my bubble
- — Cedilha no Fedora 25
- — The Woman the Mercury Astronauts Couldn’t Do Without
- — New stuff coming in C# 7.0
- — Millie 0.9.6, or “installers are hard”
- — I’m sad today
- — Une lettre ouverte à la madame qui a pointé son doigt vers ma fille
- — Philae Lander signing off
- — Best icecream I’ve ever had
- — Is it really as bad as it looks?
- — Ubuntu on Windows
- — How do I add an unknown attribute to an element in ReactJS?
- — Clashing method names in Go interfaces
- — Interfaces in Go and C#
- — Github’s Swift Style guide
- — Invincible breaks my heart
- — Simplicity is Complicated – The dot Post
- — The Evolutionary Psychology Nonsense
- — Avoiding Avoidance Behaviour
- — Why do Salespeople Believe in Magic?
- — Goodbye, Mr. Spock.
- — Go 1.4 moves the runtime toward Go
- — Where an HTTP timeout actually fires
- — My social anxiety screwed me royally this week
- — Parallel benchmarks and wandering maps
- — From Kerbin to Laythe and Back
- — sync.Pool is not a cache
- — Things I wish I knew one year ago
- — The Go 1.3 linker got less clever
- — Asking the compiler what escapes
- — What Go 1.3 did to GC pauses
- — Growing stacks without segments
- — Chasing a Go latency spike
- — The two words inside an interface
- — GOPATH is not a versioning system
- — More on validating with Go
- — [ANN] Package validator
- — Small Go packages and boring APIs
- — What json.Unmarshal really does
- — Draining HTTP response bodies
- — Validation without a reflection panic
- — Reading struct tags with reflection
- — Easter Eggs in the Apollo Program
- — Coverage, TLS, and atomic swaps in Go 1.2
- — What Go 1.2 changed under my goroutines
- — Three-index slices and capacity
- — The nil interface trap
- — When a system call takes a thread
- — What a SQL pool can and cannot do
- — How the race detector knows
- — Method values remember the receiver
- — Counting allocations in Go 1.1
- — Exemplo de texto de MBA
- — Reading lines with bufio.Scanner
- — Two small Go 1.1 speedups
- — The race detector found my clever cache
- — Join us on App.Net
- — Go 1.1 and the G-M-P scheduler
- — database/sql is not a database connection
- — Why is Change So Difficult?
- — Goodbye Intel
- — How Go waits for the network
- — Reflection is a polite form of unsafe
- — The Go memory model and my broken flag
- — gofmt is an AST program
- — Linux Kernel Linked List Explained
- — Measuring and tuning Go GC pauses
- — Container changes in C++11
- — Measuring a hot stack split
- — FOR is evil or something
- — Where did that Go value escape to?
- — Rolling out your own Fusion Drive with the recovery partition
- — Not a-ok
- — How I accidentally because a domain squatter
- — Looking under a Go map
- — The day a videogame changed my life
- — Debugging slice capacity aliasing
- — What is inside a Go interface?
- — Batching calls across cgo
- — In which Rob gets app.net
- — Go 1 stop-the-world garbage collection
- — How Go 1 goroutine stacks grow
- — Inside Go interfaces, slices, and maps
- — What Go 1 compatibility buys
- — GOPATH is a workspace, not a package
- — Go 1 and the go command
- — One tool for Go packages
- — Learning the new time package
- — Errors, runes, and delete in the Go 1 preview
- — Snapshots on the road to Go 1
- — gofix is not a test suite
- — Watching the Go garbage collector stop
- — The Go 1 preview and gofix
- — A look at the old Go scheduler
- — Could something like “Apollo 18” really have existed?
- — Profiling before guessing
- — Reflection needs a reason
- — My need for anonymity
- — Chrome and the new Lion full-screen behaviour
- — Jeff Greason on the near future of space exploration
- — errno and strings at the cgo border
- — “¿Por qué están todos hablando inglés?”
- — Godwin’s Law is anything but
- — Small Go interfaces
- — Anagramizer, a simple anagram solver in Go
- — Euler 9 in Go
- — How to set up Emacs on Windows
- — Living on Go weekly snapshots
- — Making the Scheduler Show Its Work
- — Reflection and a Tiny Config Decoder
- — Panic Is Not an os.Error
- — Cache Lines Beat Clever Loops
- — Replacing Signals With Small Interfaces
- — Merging Sorted Streams With Channels
- — One Thread, Two Threads
- — Counting Bits Like the Kernel
- — The Slice Is Not the Array
- — Reflection With the Lights On
- — Profiling the Program I Wrote
- — Euler 15 in Python
- — Euler 11 in Python
- — Euler 10 in Python
- — Euler 9 in C
- — Euler 9
- — Euler 6
- — Euler 5
- — Euler 4
- — Euler 3
- — Euler 8
- — Euler 7
- — Gob on the Wire, for Now
- — cgo Is a Border, Not a Bridge
- — Why can’t people buy newer cars in Argentina?
- — Installing Go Code With goinstall
- — Channel Pipelines With Back Pressure
- — Defer at the Exit
- — How can I get rid of (not so) old books?
- — Why do things take so long in Argentina?
- — Goroutines Are Not Pthreads
- — Interfaces After Qt
- — Declarations Without a Typewriter
- — Suquia Creek
- — From Qt Slots to Small Tools
- — A Decade as a Linux Pro
- — The Grumpiest Post I’ve Ever Written
- — The Delusion of Not Being Deluded
- — perf Turns Counters Into Questions
- — Go Is an Interesting Experiment
- — Backpressure Is Not an Error Message
- — Qt 4.5 Lowers Two Barriers
- — Btrfs Is Interesting, Not Yet Boring
- — KMS Moves the Screen Into the Kernel
- — ftrace Before More printk
- — The Allocator Is Part of the Program
- — Zero Copy Is Mostly Fewer Copies
- — epoll and the List That Does Not Grow
- — Event Loops Before More Threads
- — Cgroups Are Accounting Before Containment
- — ext4 After the Stable Label
- — QtWebKit, One Page, Three Worlds
- — QtConcurrent for the Small Jobs
- — Solid Hardware Without Device File Archaeology
- — Phonon and the Boring Audio Button
- — KWin Compositing Without the Circus
- — Plasma Is Not Just a Panel
- — KDE 4.1 Feels Like a Desktop Again
- — Idle Host, Busy Guest
- — Proxy Models Instead of Copied Lists
- — When Rendering Bugs Move
- — Trying Control Groups After 2.6.24
- — Plasma Needs Boring Jobs
- — Trolltech, Nokia and the Toolkit
- — Okular and the Printer Test
- — KDE 4.0 on My Desk
- — Chasing Black Windows
- — QtDBus and the Silent Reply
- — KVM Needs the Right Bottleneck
- — CFS After the First Week
- — KDE 4 Beta Is a Construction Site
- — Tickless Is Not Sleepless
- — KWin Compositing on Old Hardware
- — Dolphin and the Missing Knobs
- — Learning Graphics View the Slow Way
- — Oxygen Is Not the Desktop
- — KDE 4 Alpha on a Spare Partition
- — Qt Model/View Without the Magic
- — The Desktop Is Still a Workbench
- — One Event Loop, One Owner
- — KVM After 2.6.20
- — CMake Build Trees Are Disposable
- — Network State Is Not a Boolean
- — D-Bus 1.0 and Boring Interfaces
- — Plasma Data Before Decoration
- — Const Does Not Mean Concurrent
- — Solid Lines Around Hardware
- — Two Roads to Composited Windows
- — Phonon Is Not a Codec Collection
- — ThreadWeaver and the Size of a Job
- — Lock Before Looking
- — A Panel Made of Rectangles
- — Converting One CMake Target at a Time
- — HAL Is a Source, Not a Policy
- — Model/View Without the Fog
- — Porting the Seam, Not the Screen
- — CMake Before Coffee
- — The Timer That Froze the Window
- — What the Desktop Taught Me This Year
- — KDE 3.5 and the Value of Finish
- — Testing KDE 3.5 With Real Work
- — Debugging the Desktop From evdev to X.Org
- — Replacing the Directory Poll With inotify
- — Making a Use-After-Free Leave Tracks
- — Porting a Qt 3 Dialog to Qt 4
- — KDE 3.4 at Working Speed
- — A Stable Name for a Restless USB Disk
- — KDE Moves the History, Not Just the Code
- — Watching Linux History Move to Git
- — Git Starts With Objects
- — QObject Ownership Is Not Garbage Collection
- — A 2.6 Driver and the Trail of Evidence
- — Sleeping in the Kernel Without Regret
- — The Boring Desktop Test
- — X.Org Configuration With Less Guessing
- — Following the Device Through sysfs
- — Qt 3.3 Signals Without Magic
- — KDE 3.3 Feels Settled
- — Watching Latency on Linux 2.6
- — Small KDE 3.2 Habits That Save Time
- — A Model Pattern for Qt 3 List Views
- — Moving from XFree86 to X.Org
- — Flicker-Free Custom Painting in Qt 3
- — Letting udev Name the Devices
- — Moving a Real Machine from Linux 2.4 to 2.6
- — Why XFree86 Is Being Replaced
- — Threads in Userspace and the Kernel
- — What Kernel Preemption Actually Does
- — KDE 3.2 in Daily Use
- — sysfs Is Not a New proc
- — Linux 2.6.0 Arrives
- — Trying the KDE 3.2 Beta
- — Reading the Device Tree in sysfs
- — NPTL and the Disappearing Processes
- — Qt Layouts Before Pixel Pushing
- — Why the O(1) Scheduler Feels Different
- — Getting Ready for Linux 2.6-test1
- — What Safari Means for KHTML
- — Keeping Work Out of a Network Interrupt
- — Merging KParts Actions Without Duplicates
- — Building a Tiny KIO Slave
- — Let DCOP Tell You the Signature
- — Moving a KDE 3 Application to 3.1
- — Module Parameters in Linux 2.4
- — Reading an skb Without Regretting It
- — Building a Module for the Running Kernel
- — Following a File Through the Linux 2.4 VFS
- — Waking a Process in Linux 2.4
- — A Linux 2.4 Module with a Clean Exit
- — Adding a Source File to KDE Autotools
- — KDEPrint and CUPS: Where the Options Go
- — Testing KHTML with a Small Page
- — Do Not Block the KIO Job
- — KParts Ownership in One Minute
- — A Small DCOP Command Is a Good Test
- — Porting an Application to KDE 3
- — What moc Actually Generates
- — GCC 2.95 Is Part of the Platform