Re: Contexts

[Frank Heckenbach]

Bruno Haible wrote:

> > I was addressing the fact that token names might be too short
> > and be in conflict with other occurrences of the same word.
> 
> Ah yes, for this issue, contexts are a solution.
> (Now that you state what problem you want to solve, I agree with you :).)

My fault, I stated the thread with several (related) issues.

Contexts might help, but AIUI this would mean translating every
token name twice (ATM, I think Bison would need two contexts).
In de, e.g., as I said, most token names would in fact be the same
in both "contexts" (cases), so it would be a lot of redundant work
for translators, wouldn't it? Or wouldn't they mind because their
tooling makes that easy?

Akim Demaille wrote:

> The real good option I see (to get proper sentences rather than
> forms) is the one I mentioned in my message for Bison 3.8.

IIUIC, this means having to translate an individual message for each
state. That's surely a lot of work, and a small change in the
grammar can cause many messages to requires changes in each
translation, can't it? (Any many of those changes would be subtle,
requiring especially much attention of the translators.) It would
probably give the best possible result, but is it worth that huge
effort?

> Meanwhile, quotes around %s could improve the result.

I suppose you mean as a way to avoid the gender. I don't think it
works too well in de, especially if the token is not a literal
keyword or such, but a description, such as:

  unerwartetes "Buchstabe"

Nah, not without a leading term such as "unexpected token \"%s\"".

> Or, in the
> case of French, defensive translation:
> 
> > foo.y:2.1: erreur: fin de fichier inattendu(e)

This would already get ugly in de ("unerwartete(r/s)"), and in other
languages it might be more than a (few) suffix(es). I think in Latin
it might be "inopinat{us,a,um}"(sg.) or "inopinat{i,ae,a}"(pl.)
which reads more like a school book lesson than an actual message. ;)

Regards,
Frank