On Sun, 2009-10-18 at 14:17 +0200, Paolo Bonzini wrote: > On 10/18/2009 06:25 AM, Jamie Lokier wrote: > > The manual for GNU gettext explains quite well why gettext takes a > > message string as argument, instead of a "message code". Imho, a > > similar case can be made for error messages at call sites. > > That's true. However here we have the case of having errors consumed by > programs as well as users, so we want something that can be easily made > into language bindings. > > In other words, this situation is much more similar to errno/strerror, > than to a compiler error message (which will often be used by other > programs, but where the actual error text will be read by a person; this > can and should use gettext).
So to extend your analogy, I think we're reaching the conclusion that both errno *and* string should be returned to the client, right? The client can localize based on errno, and in addition optionally provide the server-provided string to the user. -- Hollis Blanchard IBM Linux Technology Center
