I like static site generators: you need less resources to host your sites (even one of the almost useless EC2’s micro instances could be enough) and you can move them around more easily, even to other hosting services with very little fuss. Obviously, there are some inherent limitations, but some of them can be overcome easily by using external services, while others cannot be reliably defeated without help from the hosting platform.

An example which intrigued me recently: your site is translated in multiple languages and the users should be offered the “best” language.

What’s the “best” language? The official one spoken in the users’ countries or the system/browser language chosen by the users themselves?

In the first case, we should try to locate the users via some geolocation mechanism, but is it really the right way to guess the users’ languages? I mean, I live in Italy but for many years (and to some extent still these days) I haven’t used Italian as the primary language of my computers; a good friend of mine lives, at the moment, in a German-speaking country but I don’t think he’s using his system in German. No, I think geolocation should not be used to guess the users’ languages, but it can be useful if your site’s contents are more tailored to countries than languages.

What’s left? Asking the users/browsers. Enter the Accept-Language header.

Good browsers send requests with the Accept-Language header listing the preferred languages, each with an optional priority values, and usually ordered according to these priorities. Users can modify this list at system or browser level (setting the system language is usually enough), so assuming it reflects the languages our users would like to use is a safer bet, I think.

Forgetting (unportable? unreliable? impossible?) attempts to get this list via JavaScript, we must rely on the web server because it’s the only “active” part of our system. My preferred one is Nginx, at the moment; to have it parse the Accept-Language header we must use a third-party module, the Nginx Accept Language module (adding it is not really difficult and thanks to a smart guy I’ve learnt a couple of ways to automate the process).

The module’s doc already shows how to use the detected language value but here are my two cents:

In this way, every generic, “language-less” request will have a chance to lead users to a translated resource, but, assuming internal links are correctly namespaced, requests for already translated resources won’t force another language upon our users.

So, with a little help from the web server we can try to guess users’ preferred languages; we’re left with problems like how to organize the static site with minimal duplication and how to reduce the effort to keep the different languages in sync, but here Nginx cannot help us :-)