Re: Hashing, and Ted's concerns about wastefulness

Graham writes, quoting Bill:
> At 15:16 02/02/99 -0800, Bill Newman wrote:
> >Ted wrote (quoting Graham)
> >
> >> Yes, but we need to specify exactly how that is represented.  Remember
> >> that most hashes on files care about every bit of whitespace and every
> >> control character; if we don't specify the representation very well,
> >> we will end up with conflicting hashes on equivalent feature expressions.
> >
> >How bad a problem is this? Is it is worth trying to assign the same
> >hash values to differently-represented-but-ultimately-equivalent
> >feature sets? It is a feature, but it adds complexity to the
> >specification and to each implementation.
> 
> Good question.
> 
> I don't have a clear answer, but I *suspect* that
> "differently-represented-but-ultimately-equivalent feature sets" won't be a
> big issue.  I have an unsubstantiated view that automatically recognizing
> the same feature set from different authors will not be a big requirement.
> >Do you anticipate a pattern of usage where this feature would 
> >significantly increase efficiency?
> 
> Personally, no.
> 
> Also, even if we did not require the hash to be computed over a fully
> canonicalized value, a combination of whitespace normalization and
> presentation conventions might still match a good number of separately
> authored cases.

Remember that typical hash functions care about control characters, so
we inherit the whole CR vs. CRLF kinds of problems.  Normalizing it now
is worth the time, not so much for the "seperately authored" case, but
so that the internal representations used to generate the hashes are
consistent across platforms with different conventions.  


				regards,
					Ted