Deprecate Xapian::weight in favour of double?

[Olly Betts]

For Xapian::docid, etc having dedicated types is definitely good - it allows rebuilding to allow more than 4 billion documents, as well as side-stepping portability issues with integer type sizes.

But for weights, it doesn't make much sense to rebuild with Xapian::weight being anything other than double. You don't need the extra precision of long doubles, while floats are likely to be the same speed or slower on anything vaguely modern, and probably a bit low precision.

I brought this up on IRC and Richard agreed it seemed sensible.

One thing which might shoot down this idea - if you can't subclass and reimplement a virtual method using Xapian::weight where the parent class uses double, it's going to be hard to provide a smooth migration path.

[Olly Betts]

The subclassing thing I mentioned works fine.

[Olly Betts]

Done in trunk:

• r16207 for xapian-core
• r16208 for xapian-omega
• r16209 for xapian-bindings