Nicolas

Have you noticed that you are the only real bottleneck in agentic coding?

Common mistake: guessing hotspots in Go. Don't micro-optimize random code. Use pprof (net/http/pprof), capture a CPU profile, inspect top/callgraph. Add expvar counters for throughput/latency. Optimize what actually burns CPU.

Common mistake in Go testing: treating coverage as proof of correctness. 100% coverage only means lines executed. Add boundary and negative cases (empty, first, last, middle, not-found) via table-driven tests, then use cover profiles to find gaps.

@Oleksj1366220 I've recently learned about TRMNL, thought it was a cool idea and remembered I had an epaper display laying around. So I naturally thought 'how hard could it be to build my own display?' 🤣

@nzachow_ It’s really cool to meet someone who builds purely to sharpen their skills. That’s pretty rare these days! Would love to hear more about what sparked the idea if you’re up for sharing.

Epaper displays are getting easier to use 🚀

When writing Go tests, use t.Helper() for assertion/fixture helpers to report failures at the caller’s file:line for faster debugging. Add t.Logf for context (IDs, inputs) and run with 'go test -v'. Filter with '-run' to isolate failing cases.

My custom dashboard is taking shape

@iannuttall In my experience, your prompt should be simple to allow the model to one-shot it. Quality degrades the longer the conversation gets.

Hot take 🌶️ Most of the anti-Claude sentiment is likely too much context bloat and switching to Codex removes that so it feels a lot “smarter” then slowly as you add MCP and prompts, Codex gets “dumber” and the cycle continues…

Profile Go benchmarks to identify performance hotspots. Run 'go test -bench=. -cpuprofile=cpu.out', then analyze with 'go tool pprof -top' or '-list'. Optimize the main time-consuming sections, not rarely executed paths.

Any ideas on how I could use this gadget? It's an esp32 + e-paper screen with a couple sensors.

In Go, organize packages by dependency direction. Place domain types and logic in pkg/domain without I/O. Have repositories and clients depend on domain. Make transports (HTTP/gRPC) depend on services. Wire dependencies in cmd/<app>/main. Use internal/ to encapsulate adapters.

Go modules allow flexible directory structures, removing GOPATH constraints. After cloning a repository, build from the module root containing go.mod to avoid import path errors. Alternatively, run `go install ./cmd/foo@latest` from any directory.

In Go, define small interfaces from the consumer's perspective. Functions should accept interfaces to specify behavior and return concrete types. This minimizes coupling and ensures substitutability.

In Go, for GC-friendly map tile caches, use sharded LRU to reduce lock contention, apply size-aware admission to exclude oversized tiles, and implement asynchronous batched evictions.

Go pprof enables scoped CPU profiles via labels. Employ runtime/pprof.Labels and pprof.Do to tag targeted workloads. Filter with go tool pprof -tagfocus=<label>. Maintain low-cardinality label values to minimize overhead.

Go pprof CPU profiler samples at approximately 100 Hz. Increasing the rate via runtime.SetCPUProfileRate adds overhead and may distort scheduling and garbage collection. For short bursts, extend profile duration or perform multiple captures.

Pro tip: Design fan-in as func FanIn(ctx context.Context, ins ...<-chan T) <-chan T. Spawn one forwarder per input: for v := range in { select { case out <- v: case <-ctx.Done(): return } }. Directional chans prevent misuse; ctx avoids leaks if the consumer stops.

CPU profiling with Go pprof collects samples from OS threads. Avoid profiling on loaded or thermally throttled systems, as data may be unreliable. Use an idle host with fixed CPU frequency and perform repeated before-and-after runs for consistent results.

Pro tip: in Go, detecting goroutine leaks by checking runtime.NumGoroutine in tests is unreliable due to noise and race conditions. Instead, ensure goroutines respond to context cancellation via ctx.Done, synchronize with a WaitGroup and timeout.

In Go testing, t.Skip terminates the test but executes deferred functions. To avoid resource waste from late invocation after setup or during cleanup, verify preconditions like environment, network, and binaries early and skip immediately.