On Fri, 15 Dec 2006, Paolo Bonzini wrote: > Joel E. Denny wrote: > > On Thu, 14 Dec 2006, Paul Eggert wrote: > > > > > Seriously, though, this business about "experimental" is kind of > > > demoralizing for users who are reading the manual. Why should they > > > invest their work in an "experimental" feature? > > > > The C++ skeletons and %destructor seemed to do okay this way, but I don't > > know the full history there. > > Yes, but I also think that we should not exaggerate.
Exaggerate what? That they did okay? Or the instability of our current directives? > Personally, I would not label as experimental even the requires/provides part > (though I do find it a bit unnerving that we need %code-top...), as well as > %language. If I had to choose only one new feature to label as experimental, it's the prologue alternatives because they keep evolving. After my recent discussion with Hans Aberg and Jeff Inman, I have yet another revision for them. (I'm thinking it'll ease all the concerns I've heard so far... including the dislike for %code-top that even I feel. However, I'm putting that off until the other current issues settle down.) Anyway, I just don't see the harm in labeling things experimental in a test release. And I believe we do need at least one test release before the next stable release. > If it wasn't for the language conventions, I would have kept the ".tab" even > in the Java output, as I don't like to depart too much from the traditional > Bison behavior. My feeling is that the community has already departed... look at Bison's own parse-gram.c and parse-gram.h.
