More consistent array traversal semantics

[judofyr]

Array traversals aren't yet defined in the spec (see #2), but the intended semantics have been implemented in e.g. groq-js. As I've been thinking and implementing these semantics I've found that they are a bit inconsistent and hard to work with. This issue is intending to make them more consistent and easy to implement.

These are the new proposed rules:

• Every expression is (at a syntactical level) has a traverse behavior which is either none, implicit or explicit.
• * and array constructor ([…]) have explicit traverse behavior .
• The postfix projection operator (base[]) works as follows:
  • If the base has no or implicit traversal behavior, then it converts it to explicit traversal behavior.
  • If the base has explicit traversal behavior, then it applies a flatten operation.
• The grouping expression ((…)) has no traverse behavior, regardless of what's inside it.
• Filtering (base[foo > 1]) and slicing (base[1..5]) preserves the traverse behavior of the base.
• Attribute access (base.foo), dereferencing (base->) and projection (base{bar}) have the following semantics:
  • If the base has either implicit or explicit traverse behavior, then it's being applied for each element in the base (e.g. *{bar} will project {bar} on each element). The returned expression has the same implicit/explicit traverse behavior as the base.
  • If the base has no traverse behavior, then it's being applied directly to the base.
• Element access (base[5]) has the following semantics:
  • If the base has explicit traverse behavior, then it's being applied for each element in the base.
  • Otherwise, then it's being applied directly to the base while preserving the traverse behavior.
• All other expressions have no traverse behavior.

Implications:

• *[_type == "bar"]{baz} still works as expected
• *[_type == "bar"]{users[]{name}}: The [] is needed when projecting over arrays.
• Nested traversals still work as expected: *[_type == "bar"]{"avatars": users[].avatar.url}

Implementations:

• I've updated the groq-test-suite in the tweak-array-traversal branch: sanity-io/groq-test-suite@sanity-io:5af96a8...sanity-io:e63e5b6. There was only one change required, but note that it doesn't have the best coverage for these cases.
• I've implemented these in groq-js in the tweak-array-traversal branch: sanity-io/groq-js@sanity-io:10a2af2...sanity-io:14fc014

[judofyr]

Closed in favor of #42.