[jdom-interest] performance on tonights CVS

[Jason Hunter]

Alex Rosen wrote:
> 
> > Ideas are most welcome, but if nothing surfaces I'm tempted to remove
> > that verification check.  For the price, I think it provides minimal
> > benefit.
> 
> Amen! The requirement that it be impossible to build a non-well-formed
> document always seemed... overly nice, to me. It's great to help developers
> out by catching errors for them when it's cheap, but not when it's
> expensive. It's not fair that you have to pay the price, even if your code
> does the right thing.
> 
> What if we didn't call the Verifier anywhere, but we added checkElement()
> and checkDocument() to it? That way a developer could optionally check their
> document, after they created it in-memory. If you're parsing a file (using a
> parser that does the proper checks), or if you're building a document
> in-memory but can guarantee that you're only adding legal values (e.g. if
> all the data you added were integers), then you wouldn't have to pay the
> price. We could move all non-free checks there, such as AttributeList's
> checks for duplicate attributes. (Maybe that's cheap enough to leave as-is,
> dunno.)
> 
> Alex

That's another alternative, and one I've considered.  But it's like
malloc returning null; few bother to check.

-jh-