Multiple canonical relation type

[YoucTagh]

In both Example 19 and Example 20, there is a Link header with two canonical relation type. Is this something acceptable?

The guidelines in RFC that introduce this relation type suggest that it is not:

- Specify only one canonical link relation for a resource.

[YoucTagh]

This comment is a summary of the discussion @youctagh, @larsgsvensson and @rob-metalinkage had on this issue during the meeting on 09 August 2023.

Problem:

In a response using the Link header, we need a way to express the default representation as well as the alternative representations. In addition, we need to express the default or preferred representation within a dimension or combination of dimensions, not just media type and profile (as mentioned in Section 7.3.2.2

... other content negotiation dimensions (language etc.)

also in Section 7.3.4

... indicate Media Type and other content negotiation dimensions.

Example:

Suppose we have a resource A identified by http://example.org/resource/a that has six representations (r_i) that vary along three dimensions: media type (m), language (l), and profile (p):

• The first representation r₁ has media type m₁, language l₁ and conforms to profile₁ (identified by http://example.org/resource/a?m=m1&p=p1&l=l1).
• r₂: m₁, l₃, p₂ (identified by http://example.org/resource/a?m=m1&p=p2&l=l3)
• r₃: m₂, l₂, p₁ (identified by http://example.org/resource/a?m=m2&p=p1&l=l2)
• r₄: m₃, l₂ (profile is unknown) (identified by http://example.org/resource/a?m=m3&l=l2)
• r₅: m₃ (language and profile are unknown) (identified by http://example.org/resource/a?m=m3)
• r₆: m₁, l₂, p₁ (identified byhttp://example.org/resource/a?m=m1&p=p1&l=l2)

We want to express that:

• r₁ is the global default representation (if no default representation is defined for a dimension or combination of dimensions).
• r₁ is the preferred representation for media type m₁ (i.e. if the request only specifies a preference for media type m₁, then between r₁, r₂ and r₆ that have this media type, r₁ is the preferred one).
• r₃ is the preferred representation for the profile p₁ (i.e. between r₁, r₃ and r₆, r₃ is the preferred one).
• r₆ is the preferred representation when the combination m₁ and p₁ is requested.
• r₃ is the preferred representation when the combination p₁ and l₂ is requested.

Proposed solution:

Use the relation types canonical and alternate (check IANA Link Relations). The former is used to define the global default representation, the latter to define an alternative representation that is neither a global default nor a dimension specific preferred representation (in our example, r₂ and r₃ and r₄). We introduce a new relation type preferred to define the preferred representation for a dimension or combination of dimensions, which is used with the new attribute dim to specify the dimension of variability.

The Link headers to express the example are:

Link:
    <http://example.org/resource/a?m=m1&p=p1&l=l1>;
        rel="canonical";
        type="m1";        format="p1";        lang="l1",

    <http://example.org/resource/a?m=m1&p=p1&l=l1>;
        rel="preferred";
        dim="m";
        type="m1";        format="p1";        lang="l1",

    <http://example.org/resource/a?m=m1&p=p2&l=l3>;
        rel="alternate";
        type="m1";        format="p2";        lang="l3",

    <http://example.org/resource/a?m=m2&p=p1&l=l2>;
        rel="preferred";
        dim="p";
        type="m2";        format="p1";        lang="l2",

    <http://example.org/resource/a?m=m3&l=l2>;
        rel="alternate";
        type="m3";        lang="l2",

    <http://example.org/resource/a?m=m3>;
        rel="alternate";
        type="m3",

    <http://example.org/resource/a?m=m1&p=p1&l=l2>;
        rel="preferred";
        dim="m p";
        type="m1";        format="p1";        lang="l2",

    <http://example.org/resource/a?m=m2&p=p1&l=l2>;
        rel="preferred";
        dim="p l";
        type="m2";        format="p1";        lang="l2"

Notes

Note that the solution to this problem may also affect other examples such as Examples 22, 23 and 24.

[rob-metalinkage]

To do:

remove multiple canonical tags from example.

add or refer to text to define precedence over dimensions.

No plans to implement a more nuanced "prefered" tag, so this will determine the degree of expressivity available.

[larsgsvensson]

I reopen and tag as "due-for-closing"

[rob-metalinkage]

Resolved in PR #46