The situation with Go on OpenBSD
Ouch. @cks himself isn't making that claim, but it seems to me that anything written in Go is unusable on OpenBSD.
I don't think the situation is usually that bad. Our situation with Prometheus is basically a worst case scenario for Go on OpenBSD, and most people will have much better results, especially if you stick to supported OpenBSD versions.
If you stick to supported OpenBSD versions, upgrading your machines as older OpenBSD releases fall out of support (as the OpenBSD people want you to do), you should not have any problems with your own Go programs. The latest Go release will support the currently supported OpenBSD versions (as long as OpenBSD remains a supported platform for Go), and the Go 1.0 compatibility guarantee means that you can always rebuild your current Go programs with newer versions of Go. You might have problems with compiled binaries that you don't want to rebuild, but my understanding is that this is the case for OpenBSD in general; it doesn't guarantee a stable ABI even for C programs (cf). If you use OpenBSD, you have to be prepared to rebuild your code after OpenBSD upgrades regardless of what language it's written in.
(You may have to change the code too, since OpenBSD doesn't guarantee API compatibility across versions either. But the API is generally more stable than the ABI.)
If you have older, out of support OpenBSD versions and Go programs that you keep developing, you're fine as long as you make your Go code work with the oldest Go release that's required for your oldest OpenBSD. This mostly means avoiding new additions to the Go standard library, although sometimes performance improvements make new patterns in Go code more feasible and you'll have to not rely on them. You'll probably have to freeze and maintain your own binary copies of old Go releases for appropriate old OpenBSD versions. Go's currently ongoing switch from its old way of just fetching packages to modules may cause you some heartburn, but you can probably deal with this by vendoring everything.
(If you have Go programs that you don't keep developing, life is even easier because you can freeze your pre-build binaries for older OpenBSD versions as well. All you need to do is periodically build the latest Go for the latest OpenBSD and then build your programs with it.)
If you have older OpenBSD releases as well as current ones, and a Go program where you want to use the latest Go features (or where you want to keep up with dependencies and those dependencies do), you can still do this provided that you don't need to run new versions of your program on old OpenBSDs. If they're fine with the version they have, even if it's not as efficient or as feature-full as the latest one, then you can just freeze their pre-build binaries and only run the latest version of your program on the latest OpenBSDs with the latest Go.
And this gets us down to Prometheus, where we have the worst case scenario; you need to run the current version of your program on all OpenBSD versions and the current version uses features from very recent versions of Go, which only run on a few OpenBSD versions. This doesn't work, as discussed.
Sidebar: Cgo and OpenBSD
In general, avoiding cgo is likely to extend the range of OpenBSD versions that a Go program can run on, even a Go program that's built with the very latest version of Go. Although older versions of OpenBSD aren't officially supported by Go 1.14, it can cross-build a pure Go program that seems to work as far back as OpenBSD 5.8 (on 64-bit x86), although it fails on OpenBSD 5.4 with a 'bad system call' core dump. Using cgo gives you two problems; you can't cross-build, and you're likely to be much more restricted in what range of OpenBSD versions the resulting binaries will run on.
(Being able to cross-build from Linux makes it much easier to set up older Go versions to build binaries for old OpenBSD versions, since old Go versions will generally build fine even on new Linuxes.)