3.2.2. Views Collation

3.2.2.1. Basics

View functions specify a key and a value to be returned for each row. CouchDB collates the view rows by this key. In the following example, the LastName property serves as the key, thus the result will be sorted by LastName:

  1. function(doc) {
  2. if (doc.Type == "customer") {
  3. emit(doc.LastName, {FirstName: doc.FirstName, Address: doc.Address});
  4. }
  5. }

CouchDB allows arbitrary JSON structures to be used as keys. You can use JSON arrays as keys for fine-grained control over sorting and grouping.

3.2.2.2. Examples

The following clever trick would return both customer and order documents. The key is composed of a customer _id and a sorting token. Because the key for order documents begins with the _id of a customer document, all the orders will be sorted by customer. Because the sorting token for customers is lower than the token for orders, the customer document will come before the associated orders. The values 0 and 1 for the sorting token are arbitrary.

  1. function(doc) {
  2. if (doc.Type == "customer") {
  3. emit([doc._id, 0], null);
  4. } else if (doc.Type == "order") {
  5. emit([doc.customer_id, 1], null);
  6. }
  7. }

To list a specific customer with _id XYZ, and all of that customer’s orders, limit the startkey and endkey ranges to cover only documents for that customer’s _id:

  1. startkey=["XYZ"]&endkey=["XYZ", {}]

It is not recommended to emit the document itself in the view. Instead, to include the bodies of the documents when requesting the view, request the view with ?include_docs=true.

3.2.2.3. Sorting by Dates

It maybe be convenient to store date attributes in a human readable format (i.e. as a string), but still sort by date. This can be done by converting the date to a number in the emit() function. For example, given a document with a created_at attribute of 'Wed Jul 23 16:29:21 +0100 2013', the following emit function would sort by date:

  1. emit(Date.parse(doc.created_at).getTime(), null);

Alternatively, if you use a date format which sorts lexicographically, such as "2013/06/09 13:52:11 +0000" you can just

  1. emit(doc.created_at, null);

and avoid the conversion. As a bonus, this date format is compatible with the JavaScript date parser, so you can use new Date(doc.created_at) in your client side JavaScript to make date sorting easy in the browser.

3.2.2.4. String Ranges

If you need start and end keys that encompass every string with a given prefix, it is better to use a high value Unicode character, than to use a 'ZZZZ' suffix.

That is, rather than:

  1. startkey="abc"&endkey="abcZZZZZZZZZ"

You should use:

  1. startkey="abc"&endkey="abc\ufff0"

3.2.2.5. Collation Specification

This section is based on the view_collation function in view_collation.js:

  1. // special values sort before all other types
  2. null
  3. false
  4. true
  5. // then numbers
  6. 1
  7. 2
  8. 3.0
  9. 4
  10. // then text, case sensitive
  11. "a"
  12. "A"
  13. "aa"
  14. "b"
  15. "B"
  16. "ba"
  17. "bb"
  18. // then arrays. compared element by element until different.
  19. // Longer arrays sort after their prefixes
  20. ["a"]
  21. ["b"]
  22. ["b","c"]
  23. ["b","c", "a"]
  24. ["b","d"]
  25. ["b","d", "e"]
  26. // then object, compares each key value in the list until different.
  27. // larger objects sort after their subset objects.
  28. {a:1}
  29. {a:2}
  30. {b:1}
  31. {b:2}
  32. {b:2, a:1} // Member order does matter for collation.
  33. // CouchDB preserves member order
  34. // but doesn't require that clients will.
  35. // this test might fail if used with a js engine
  36. // that doesn't preserve order
  37. {b:2, c:2}

Comparison of strings is done using ICU which implements the Unicode Collation Algorithm, giving a dictionary sorting of keys. This can give surprising results if you were expecting ASCII ordering. Note that:

  • All symbols sort before numbers and letters (even the “high” symbols like tilde, 0x7e)
  • Differing sequences of letters are compared without regard to case, so a < aa but also A < aa and a < AA
  • Identical sequences of letters are compared with regard to case, with lowercase before uppercase, so a < A

You can demonstrate the collation sequence for 7-bit ASCII characters like this:

  1. require 'rubygems'
  2. require 'restclient'
  3. require 'json'
  4. DB="http://127.0.0.1:5984/collator"
  5. RestClient.delete DB rescue nil
  6. RestClient.put "#{DB}",""
  7. (32..126).each do |c|
  8. RestClient.put "#{DB}/#{c.to_s(16)}", {"x"=>c.chr}.to_json
  9. end
  10. RestClient.put "#{DB}/_design/test", <<EOS
  11. {
  12. "views":{
  13. "one":{
  14. "map":"function (doc) { emit(doc.x,null); }"
  15. }
  16. }
  17. }
  18. EOS
  19. puts RestClient.get("#{DB}/_design/test/_view/one")

This shows the collation sequence to be:

  1. ` ^ _ - , ; : ! ? . ' " ( ) [ ] { } @ * / \ & # % + < = > | ~ $ 0 1 2 3 4 5 6 7 8 9
  2. a A b B c C d D e E f F g G h H i I j J k K l L m M n N o O p P q Q r R s S t T u U v V w W x X y Y z Z

3.2.2.5.1. Key ranges

Take special care when querying key ranges. For example: the query:

  1. startkey="Abc"&endkey="AbcZZZZ"

will match “ABC” and “abc1”, but not “abc”. This is because UCA sorts as:

  1. abc < Abc < ABC < abc1 < AbcZZZZZ

For most applications, to avoid problems you should lowercase the startkey:

  1. startkey="abc"&endkey="abcZZZZZZZZ"

will match all keys starting with [aA][bB][cC]

3.2.2.5.2. Complex keys

The query startkey=["foo"]&endkey=["foo",{}] will match most array keys with “foo” in the first element, such as ["foo","bar"] and ["foo",["bar","baz"]]. However it will not match ["foo",{"an":"object"}]

3.2.2.6. _all_docs

The _all_docs view is a special case because it uses ASCII collation for doc ids, not UCA:

  1. startkey="_design/"&endkey="_design/ZZZZZZZZ"

will not find _design/abc because ‘Z’ comes before ‘a’ in the ASCII sequence. A better solution is:

  1. startkey="_design/"&endkey="_design0"

3.2.2.7. Raw collation

To squeeze a little more performance out of views, you can specify "options":{"collation":"raw"} within the view definition for native Erlang collation, especially if you don’t require UCA. This gives a different collation sequence:

  1. 1
  2. false
  3. null
  4. true
  5. {"a":"a"},
  6. ["a"]
  7. "a"

Beware that {} is no longer a suitable “high” key sentinel value. Use a string like "\ufff0" instead.