I think we can fix this issue by changing the links on the https://golang.org/dl/ HTML page to use https://golang.org/dl/{file}. That will help establish them as canonical URLs, and make it less troublesome if we ever need to change where the actual bytes are served from. If someone wants to be able to do curl without -L flag, they can continue to use the redirect target.

Am I missing any context? Thoughts?

IIRC the original point of that handler was just to make go get golang.org/dl/go1.n.m work (as in https://twitter.com/golang/status/1248018480479096833 etc).

But once we had the go-get metadata, we needed to do something the URLs about non-cmd/go clients. That was never the goal, though.

In 2015, getHandler was serving the list of downloads (the https://golang.org/dl/ page), the files themselves (via redirect to a GCS URL at the time), and 404 for everything else. This is from CL 14842:

func getHandler(w http.ResponseWriter, r *http.Request) {
	name := strings.TrimPrefix(r.URL.Path, "/dl/")
	if name == "" {
		listHandler(w, r)
		return
	}
	if !fileRe.MatchString(name) {
		http.NotFound(w, r)
		return
	}
	http.Redirect(w, r, gcsBaseURL+name, http.StatusFound)
}

The ?go-get=1 code path for golang.org/dl/go1.n.m commands was added in 2018 via CL 123679.

There haven't been concerns or objections raised regarding the solution I suggested in #38713 (comment). I plan to proceed with it soon. In the unlikely case we learn of something that goes poorly as a result, this will be easy to undo.

Change https://golang.org/cl/238517 mentions this issue: internal/dl: use /dl/{file} URLs on /dl/ HTML page