Test for the immutable attribute

[xtuc]

We are considering sending Cache-Control: immutable... for files on cdnjs.com (cdnjs/cdnjs#8655). However, the browser support is not ideal and it's hard to understand what's going to happen.

It would be very helpful to us to test the immutable attribute as part of the cache tests and I would be happy to invest the time. What do you think?

[mnot]

That would be great!

The tricky thing is to put the browsers into a state where they're reloading, to trigger the semantics of immutable. I think that could be achieved by using a cache mode of no-cache on the Fetch call, but it'd be important to verify that it has the desired effect on each browser by manually observing what's happening on the wire.

Happy to help / review / etc. as you like.

[ericlaw1979]

I have some tests and discussion at https://bayden.com/test/cache/refresh.aspx. One challenge is that there are so many different sorts of "reload" in different browsers. I think moving to fetch has improved the situation somewhat, but browsers are not all in perfect alignment.

[mnot]

Hm. A quick test shows that none of the major 3 browser engines seems to honour immutable from fetch with a cache mode of no-cache.

My reading of the fetch spec's definition of those semantics (scroll down) is that they arguably should, or at least that it would make sense to; i.e., that no-cache is effectively a reload (although one might say it's a 'forced' reload; that depends on whether you take no-cache to be a indication from the client that a reload is necessary, but I'd say the major use case for that is corrupted or incomplete data, in which case reload would be the right directive).

@annevk and @mcmanus, comments?

[annevk]

See discussion in whatwg/fetch#308.

[annevk]

(I would love it if we could get to a point where new "protocol" features have some kind of end-to-end integration performed so that everyone is aligned on what should happen everywhere in the pipeline before we declare it ready. That might help reduce the amount of time we spent on fixing things many years into the future.)

[ericlaw1979]

none of the major 3 browser engines seems to honour
immutable from fetch with a cache mode of no-cache.

Do they all behave the same way in terms of request headers? E.g. do the requests send the same Cache-Control request header, and do they send an If-None-Match for the previously-cached resource?

[mnot]

Looks like they are (max-age=0, as Fetch requires). However, note whatwg/fetch#722 .

[mnot]

AFAICT we're testing this, it's just that the browsers aren't following the spec. Please add new information if you find it; closing for now.