We replaced a 40 MB HTTP client with 600 lines
The hard part was not the requests. It was the error handling everyone had forgotten about.
The old client pulled in a tree of packages for retries, proxies and telemetry. Most of that tree was disabled in production.
The replacement handles timeouts, redirects and a small retry policy. It deliberately does not stream multipart uploads, which is the one feature we kept the old tool for.
The win is not size. It is that a new engineer can read the whole thing in an afternoon.
More in this section
🔒The quiet return of single-file binaries
Static linking is fashionable again. The reasons are less about performance and more about deployment fatigue.
A minimal CI pipeline in 200 lines
No plugins, no marketplace, no YAML sprawl. Just a runner, a queue, and honest logging.
🔒Editorial: your dependency tree is a supply chain
Treat it like one — pin, audit, mirror, and stop trusting transitive installs by default.