Re: Proposal for issue #101 (strong/weak validators).

On fre, 2008-11-14 at 23:10 +0100, Werner Baumann wrote:

> "cases where the validator is use does not allow reliable identification 
> of changes". In current practise this means changes within the same 
> second and there is nothing that restricts the kind of changes that 
> might occur.

I assume you by "current practise" mean the Apache ETag implementation?
There this is only true in very specific conditions. Most sub-second
changes do infact get properly reflected in the weak ETag. The server
application need to do some quite special things to not get the change
reflected in ETag or be very unlucky, unless ofcourse the ETag algorithm
has manually been degraded in quality (may be needed for replicated
cluster setups).

Yes, there is some cases where the simple algorithm used by Apache will
fail and emit the same weak ETag for two quite different objects, but in
real life use those is quite rare. In fact I would argue that it's
probably more likely the content gets updated while being sent, making
even their strong ETags "worthless", and the same for any server on any
OS where files may be updated while read by another application unless
you buffer the whole selected representation to calculate the ETag.

And no, it's not strictly compliant by Apache to return the same ETag
when the content is completely different, but the Apache team has
accepted this risk as the risk of this happening in any normal setup and
use is very low. In nearly all cases the kind of changes which the ETag
algorithm would miss is indeed of "weak semantic level".

Regards
Henrik