I counted the number of native word copied in WithContext function. The number is 43+.

.WithContext returns a full copy of the request + URL, so it's expected to be heavy.

I don't see much that can be done here. We are just copying the request, context and the url of the original request.

Any optimization will deviate from the original behavior, for eg. like just updating the context to the same request. And it seems that is exactly what you are doing too. You are re-assigning the new request to the old request.

@bradfitz's http client v2 addresses this problem: https://github.com/bradfitz/exp-httpclient/blob/master/problems.md#context-support-is-oddly-bolted-on

That's what I said. You cannot optimize this further without breaking original behavior. For v2, we can break behavior and possibly optimize this. Perhaps you want to add a note for optimizing context behavior in #5465 ?

I'm going to close this as unfortunate for now. We can address this in a future rewrite of the client package.

For now, if you need to make Requests with Contexts cheaply, you can have a global variable holding an http.Request zero value and then use its WithContext to make your new requests. For instance, if you wanted to make a version of net/http.NewRequest that took a context, users today can write:

var zeroRequest http.Request

func NewRequestContext(ctx context.Context, method, url string, body io.Reader) (*Request, error) {
	u, err := parseURL(url) // Just url.Parse (url is shadowed for godoc).
	if err != nil {
		return nil, err
	}
...
	req := zeroRequest.WithContext(ctx)
        req.URL = u
        req.Method = method
        ...
}

And that's pretty efficient. (no deep cloning of an existing URL)