Gettext support in OCaml

Question

I'd like to get i18n working in my OCaml code (translating text messages into the user's preferred language). Most translatable strings are the first non-labelled argument to one of a small group of functions. e.g.

• Text output: print "Feed '%s':" new_feed
• Errors: raise_safe "Local feed file '%s' does not exist" path
• Logging: log_warning ~ex "Error checking signature for %s" feed

In the GUI code, strings in ~label arguments should be extracted too.

Is there some way I can extract these strings automatically? To make things more interesting, I use some syntax extensions (e.g. Lwt) too.

I see there is an ocaml-gettext package, but:

• It only seems to support a fixed set of functions (f_, s_, etc).
• OPAM says nothing uses it.
• It failed to install (had to unset $NAME first), then failed to uninstall too, which also suggests it's not heavily used.

What do people actually use/recommend?

Answer 1

This is an issue I want to address in my Gasoline project, an application toolkit.

An answer from the future

In the examples, there is an over-engineered wordcount program, which serves as use-case for the library. The function Component.Message.cannot_open uses its format string as a key in an internationalised messages database:

let cannot_open name reason =
  send sink Error "${FILENAME:s}: cannot open (${REASON:s})"
  [ "FILENAME", make String name;
    "REASON", make String reason;
  ]

The sink is a global state of the software component it belongs to, it is bound to a database of internationalised messages where a translation is looked up and used to prepare the actual message.

Was suggested by the example, this facility supports formatting indications, much like printf does, we may write ${FILENAME:+20s} for instance.

Corrections for the present

For now, there is no actual database lookup implemented. The send function is defined in the functor Generic_message.Sink.Make parametrised by a Database module. The current implementation of wordcount uses the following implementation of the database:

module Database =
struct
  type t = unit
  type connection_token = unit
  let opendb () = ()
  let find () id = id
  let close () = ()
end

It is however possible to replace this definition with an actual database lookup — which is in the backlog.

Organisation of messages

In my opinion, message internationalisation should be accomplished by software components — the largest organisational unity below the whole application — and libraries or similar facilities should use standard constant sum types to report errors.