> > I think our decision not to keep just encoded versions around > > immunizes us from that one; I don't see how a redecoding could arise, > > as encoded versions follow different paths to encoding-accepting > > clients than decoded versions to unaccepting, purist clients. > > I do not quite follow what you are saying here. > > The issues is not about what happens within a single Squid but what > happens at the clients or in a cache mesh.
I was wrong. Yes, indeed, recodings can happen. > If you modify the ETag to include details on how the object has been > recoded then you are immune as each variant then has a different identity. > Also if you use weak etags you are mostly immune to your own actions, but > there is secondary caching implications where clients may get a different > encoding than expected because the two are told to be semantically > equivalent. So, are you suggesting that, for example, if we get an uncoded server response with ETag: "page12345", then we would tag a gzip-coded version as ETag: "gzippage12345"? Jon