Improve rel parsing

[Zegnat]

• Parsing of the rel attribute has been improved by using WHATWG’s definition of splitting on whitespace. We also treat it as a set, which means we drop duplicate values. This might spare us a loop or two.

• The properties (hreflang, media, title, type, text) we add to the rel-urls object should only be added if they weren’t previously set. Overwriting order has been changed to accord with this.

  The test testRelURLsInfoMergesCorrectly is a simple check for the correct behaviour.

• The final rels array within the rel-urls object has been changed to contain only unique items, sorted alphabetically. This reflects the latest parsing specification change.

  The test testRelURLsRelsUniqueAndSorted is a simple check for both uniqueness and order.

• The arrays of URLs within the rels object should not contain duplicates. The code has been changed to not add any URLs that are already present in the array. Based on the following line in the parsing spec, that had me reread it thrice as a non-native English speaker:

  • if url is not in the array of the key rel-value in the rels hash then add url to the array

  The test testRelURLsNoDuplicates is a simple check to make sure duplicate URLs aren’t added.

Closes #159 if merged.

[Zegnat]

I didn’t touch the alternates parsing because its behaviour isn’t in the parsing specification. I can’t comment on its correctness.

[Zegnat]

• Do not test the availability of text by converting strings to booleans. This creates a falsy test where strings like '0' will be treated as empty. Explicitly test for a string length.

  The test testRelURLsFalsyTextVSEmpty makes sure that an empty string '' and a falsy string '0' are evaluated differently.

This was sparked when looking into how attributes were parsed for microformats/microformats2-parsing#32. I have not tried to otherwise validate how well the parser implements that part of the rel spec and may have to be revisited when the spec issue is resolved.

[gRegorLove]

👍