Add context-based MIME sniffing

[IdeaHat]

Follow on issue to #3144

Context based mime sniffing is described in https://mimesniff.spec.whatwg.org/#context-specific-sniffing. Certain types (such as fonts) can only be sniffed from a context. However, this will require more metadata to be handed to the sniffer task.

[jdm]

We should add a Context enum and require it to be passed to the LoadData constructor. We can then use this in start_sending_sniffed_opt (pass it as an argument) and have special rules for each of the contexts described in the spec.

[jdm]

Code: components/net_traits/lib.rs

[Yoric]

I can try to work on that. @jdm, do I understand correctly that the Context enum would be one of Browsing/Image/AudioVideo/Plugin/Style/Script/Font/Text Track/Cache Manifest?

[jdm]

That looks fine for a first pass. We'll probably end up mapping those to the values for initiator, type, and destination in the fetch specification at some point, but let's focus on the MIME sniffing spec for now.

[Yoric]

I'll do that. Any suggestion how I can test it?

[jdm]

Using a local web server and running Servo with --sniff-mime-types will allow the snifing code to actually execute.

[jdm]

A straightforward testcase to use will be loading a URL which has a non-text/html Content-Type header but returns HTML content.

[Yoric]

Testing in progress. Do you think it would be possible to put something in the testsuite?

[Yoric]

So, I configured Apache to associate mp4 to application/video and I ran the following:

./mach run --sniff-mime-types http://localhost/~david/test/text_html_content_with_non_text_html_mime_type.mp4

where my file is actually html content.

Servo renders it as html, regardless of --sniff-mime-types, I'm not sure how I should interpret the result. In contrast, Firefox attempts to render as video.

[jdm]

Oh right, #7706 is about changing our HTML parser to actually care about the MIME type. If you use a type like image/jpeg instead, that should show a difference with --sniff-mime-types.

[Yoric]

How so? Won't Servo attempt to sniff it in a Browser context regardless of my patch?

[jdm]

http://mxr.mozilla.org/servo/source/components/script/dom/servohtmlparser.rs#115 looks at the computed mime type and creates an image document if we think we have an image. If we sniff the response first and discover HTML content, the result should be a non-image document with sniffing enabled.

[Yoric]

I still don't follow. Unless I misunderstand the specs, by default, Servo will be in Browser context, so it will sniff exactly the same thing regardless of my patch, right?