[whatwg] New URL Standard

[David Sheets]

On Tue, Sep 25, 2012 at 2:13 PM, Glenn Maynard <glenn at zewt.org> wrote:
> On Mon, Sep 24, 2012 at 7:18 PM, David Sheets <kosmo.zb at gmail.com> wrote:
>>
>> Always. The appropriate interface is (string * string?) list. Id est,
>>
>> an association list of keys and nullable values (null is
>> key-without-value and empty string is empty-value). If you prefer to
>> not use a nullable value and don't like tuple representations in JS,
>> you could use type: string list list
>>
>> i.e.
>>
>>
>> [["key_without_value"],[""],["key","value"],[],["numbers",1,2,3,4],["",""],["","",""]]
>
>
> This isn't an appropriate interface.  It's terrible for 99.9% of use cases,
> where you really want dictionary-like access.

This is the direct representation of the query string key-value convention.

Looking up keys is easy in an association list. Filtering the list
retains ordering. Appending to the list is well-defined. Folding into
a dictionary is trivial and key merging can be defined according to
the author's URL convention.

> The right approach is probably to expose the results in an object-like form,
> as Tab suggests, but to store the state internally in a list-like format,
> with modifications defined in terms of mutations to the list.

This sounds more complicated to implement while maintaining
invariants. A dictionary with an associated total order is an
association list.

> That is, parsing "a=1&b=2&a=3" would result in an internal representation
> like [('a', '1'), ('b', '2'), ('a', '3')].  When viewed from script, you see
> {a: ['1', '3'], 'b': ['2']}.  If you serialize it right back to a URL the
> internal representation is unchanged, so the original order is preserved.
> The mutation algorithms can then do their best to preserve the list as
> reasonably as they can (eg. assigning query.a = ['5', '6'] would remove all
> 'a' keys, then insert items at the location of the first removed item, or
> append if there were none).

Why hide the order?

>> Is this not already supported by creating a new URL which contains
>> only a relative query part?
>>
>> Like: query = new URL("?a=b&c=d"); query.query["a"] = "x";
>> query.toString() == "?a=x&c=d";
>>
>> Why is a new interface necessary?
>
>
> That won't work, since "?a=b&c=d" isn't a valid URL.

"?a=b&c=d" is a valid URI reference. @href="?a=b&c=d" is valid.

> The invalid flag will
> be set, so the change to .query will be a no-op, and .href (presumably what
> toString will invoke) would return the original URL, "?a=b&c=d", not
> "?a=x&c=d".  You'd need to do something like:
>
> var query = new URL("http://example.com?" + url.hash);
> query.query.a = "x";
> url.hash = query.search.slice(1); // remove the leading "?"
>
> That's awkward, but maybe it's good enough.

This is a use case for parsing without composed relative resolution.