Re: Satisfying IMS from HEAD or IMS would help where browser caches

David J Woolley wrote:

> I found the following URL which suggests that someone (not sure how
> official this is) was considering caching IMS responses in squid:
>
> http://hem.passagen.se/hno/squid/historic/IMS.html

That page are a collection of notes on how I thought IMS could be
implemented in Squid, and is written long before Squid know how to do
validations of cached objects. Not all of the ideas are implemented in
the current Squid release, and not all of the ideas implemented in Squid
are on that page either.

> Is there anything in HTTP/1.1 or common misimplementations of HTTP
> that would make satisfying IMS from HEAD unsafe (assuming reasonable
> plausibility checks are done)?

No. Satisfying IMS queries from cached HEAD replies is OK. HEAD are
guaranteed to return the same headers as a GET reply on the same object,
and IMS only requires the headers.

I am not aware of any HTTP implementations that returns different
headers on a HEAD request than on a GET request, but I am aware of a
couple where HEAD does not work.

Known misimplementations of HEAD:
a) HEAD unsupported and a error is returned.
b) HEAD returns the whole object as if GET was used.

> Would it be so unsafe as to make it unreasonable to have a leaf
> cache option to permit it? If not, doing TCP-IMS-HIT from HEAD
> for GET requests, would allow me to fetch just the HEADs for
> recent TCP-MISS/304 events.

Why not simply cache 304 replies?

---
Henrik Nordstrom
http://hem.passagen.se/hno/Welcome.html