Content-Type header parsing about http-core

[reschke]

This appears like a browser issue to me. There is and never was an intention to have different rules.

[annevk]

I'm not sure what that means. If it affects browsers, why would it not affect other clients in the HTTP ecosystem? Don't we all consume the same content?

[reschke]

Once again, there are no different rules. Browsers apparently implemented workarounds for misconfigured servers, but that doesn't mean that what browsers do is what everybody else is doing.

[mnot]

Do we need to do more than reference mimesniff (see #51)? Note there's already generic text about sniffing in Content-Type.

[annevk]

This issue is really about parsing Content-Type and determining its (parsed) value, irrespective of any sniffing that might also take place.

[annevk]

More concretely, text/html, text/plain means text/plain in browsers. text/plain; means text/plain (not an error). text/html;charset=gbk, text/html means text/html;charset=gbk (yup, charset is copied over as a special case).

whatwg/fetch#831 has a patch for Fetch to define this in detail with an acknowledgment that it doesn't match HTTP.

It seems greatly preferable to uplift this kind of thing and agree on header parsers across the HTTP ecosystem, but if that's not doable so be it.

[mnot]

Discussed in Basel; we could document a suggested error handling algorithm, as long as it's clear that rejecting the message because the content-type is invalid is still acceptable.

Suggested algorithm would be to use the last syntactically valid type. @annevk does that align with fetch?

[annevk]

More than today, but it's also more involved: https://fetch.spec.whatwg.org/#content-type-header.

[mnot]

Waiting for terminology from #193.

[reschke]

I just updated my ancient tests for content-type, see http://test.greenbytes.de/tech/tc/httpcontenttype.

To my surprise, Safari apparently handles comma-separated different from multiple fields, see http://test.greenbytes.de/tech/tc/httpcontenttype/#multitextxmloneline.

@annevk, that would be a bug according to FETCH, right?

[annevk]

Yes, I suspect there's a bug filed already. I think I found such bugs in all browsers as they all have very similar logic in their core HTTP parsers.

[reschke]

Ah. But then, I'm not really convinced that we should add this to the spec. My understanding that the request was based on browsers already agreeing on the error handling...

[annevk]

There's no complete agreement on all details involved. See whatwg/fetch#831 (comment) for links to bugs.

[reschke]

I did some more testing with Java client libs (HTTPURLConnection, JDK HttpClient, Apache HttpClient). Of these, only the first seems to have an API getting the type directly, and that indeed takes the last value (no matter whether sent as list in a single instance, or in multiple fields).

It would be nice if there was a concrete Webkit bug that points out their inconsistency (see #39 (comment)) (cc @tfpauly).

[annevk]

I recommend either filing one that blocks https://bugs.webkit.org/show_bug.cgi?id=192000 or detailing the specific case you care about there in a comment.

[reschke]

done: https://bugs.webkit.org/show_bug.cgi?id=215604