Re: *Proposed change* _data-tag values give vector length

[Erik Ostrom <eostrom@research.att.com>]

> I am in favor of punting the specification of "possible" length
> completely, as a complication that buys little.

I have little opinion for or against it.  I can see how it would aid certain 
implementations and I can see that it would add complexity, but I suspect it's 
pretty minor either way.  (In particular, given that an implementation can 
completely ignore the length specifier, the complexity added seems to be 
nothing more than an extra paragraph worth of spec to read.)

> It should not be an exact length because that would make
> significantly complex certain uses ("streaming") to which it
> being put right now, where the data is not known at the the time
> the message is sent, but is being computed on the fly.

Not true; whether a non-zero value indicates approximate length or exact 
length, a 0 can still be defined as "I don't know".  (Of course, this means 
that 0 can't be used to indicate exact length of 0, a potentially useful 
case.)  I can imagine implementations which could benefit from all three 
levels of length information:  Undetermined, approximate, and exact.  But 
again, I don't feel very strongly about it.

--Erik