Better error handling for dates

Why these changes are being introduced:

We encountered a bug after V2 launch in the normalization process,
during which Bento hits an unhandled nil value that breaks the
search results. This is because we are assuming every ASpace
record has a creation date, when in fact some of them have
publication dates instead.

Somewhat related to this is a transmogrifier bug that causes a
`kind` to be associated with a date even when there is no date.

Relevant ticket(s):

* https://mitlibraries.atlassian.net/browse/TIMX-211
* https://mitlibraries.atlassian.net/browse/TIMX-212

How this addresses that need:

This updates `NormalizeTimdex#extract_dates` to select as relevant
dates of either the 'creation' _or_ 'publication' date kind. It
also adds a guard clause to return if the `relevant_dates` array
is empty (a precaution that would have prevented this error).

Side effects of this change:

I removed a superfluous check for `date['kind'] == 'creation'` in
the guard clause in the `extract_dates` method that checks for
a date range. At this point in the method, we have already
confirmed that the relevant date is either a creation or publication
date.

Author: jazairi

app/models/normalize_timdex.rb

@@ -38,17 +38,18 @@ def extract_title(record)
   end
 
   def extract_dates(dates)
-    # It is unlikely for a record to have more than one creation date, but just in case...
-    relevant_dates = dates.select { |date| date['kind'] == 'creation' }
+    # It is unlikely for a record to have more than one creation or publication date, but just in case...
+    relevant_dates = dates.select { |date| date['kind'] == 'creation' || date['kind'] == 'publication' }
 
-    # If the record *does* have more than one creation date, it's probably not worth determining which to display.
-    return if relevant_dates.count > 1
+    # If the record has no creation or publication date, stop here.
+    return if relevant_dates.empty?
 
+    # If the record *does* have more than one creation/pub date, just take the first one.
     relevant_date = relevant_dates.first
 
-    # We are only concerned with creation dates that are ranges, since we harvest ASpace metadata at the collection
+    # We are only concerned with creation/pub dates that are ranges, since we harvest ASpace metadata at the collection
     # level.
-    return unless relevant_date['kind'] == 'creation' && relevant_date['range'].present?
+    return unless relevant_date['range'].present?
 
     "#{relevant_date['range']['gte']}-#{relevant_date['range']['lte']}"
   end

test/models/normalize_timdex_test.rb

@@ -36,4 +36,46 @@ def aspace_records
     assert_match 'The Paul Earls archives contains a large number of music manuscript',
                  aspace_records['results'][0].blurb
   end
+
+  # Some ASpace records have a publication date range rather than a creation date range. The following regression test
+  # ensures that the NormalizeTimdex model does not error for such records.
+  test 'records with no creation date do not error' do
+    VCR.use_cassette('aspace publication date',
+                     allow_playback_repeats: true) do
+      raw_query = SearchTimdex.new.search('SOVEREIGN INTIMACY: PRIVATE MEDIA AND THE TRACES OF COLONIAL VIOLENCE')
+      assert_not raw_query['data']['search']['records'].first['dates'].first['kind'] == 'creation'
+      assert_nothing_raised do
+        NormalizeTimdex.new.to_result(raw_query, 'SOVEREIGN INTIMACY: PRIVATE MEDIA AND THE TRACES OF COLONIAL VIOLENCE')
+      end
+    end
+  end
+
+  # This is another regression test. Now that we know that records with no creation date will not error, we also want
+  # to ensure that records with a publication date will be normalized accordingly.
+  test 'publication dates are normalized' do
+    VCR.use_cassette('aspace publication date',
+                     allow_playback_repeats: true) do
+      raw_query = SearchTimdex.new.search('SOVEREIGN INTIMACY: PRIVATE MEDIA AND THE TRACES OF COLONIAL VIOLENCE')
+      normalized = NormalizeTimdex.new.to_result(raw_query, 'SOVEREIGN INTIMACY: PRIVATE MEDIA AND THE TRACES OF COLONIAL VIOLENCE')
+
+      assert_equal 'publication', raw_query['data']['search']['records'].first['dates'].first['kind']
+      assert_equal '1940', raw_query['data']['search']['records'].first['dates'].first['range']['gte']
+      assert_equal '1983', raw_query['data']['search']['records'].first['dates'].first['range']['lte']
+      assert_equal '1940-1983', normalized['results'][0].year
+    end
+  end
+
+  test 'normalizer selects the first of two relevant dates' do
+    VCR.use_cassette('aspace multiple creation dates') do
+      raw_query = SearchTimdex.new.search('Timurid Architecture Research Archive')
+      assert raw_query['data']['search']['records'].first['dates'].count > 1
+      assert_not_equal raw_query['data']['search']['records'].first['dates'].first,
+                       raw_query['data']['search']['records'].first['dates'].second
+
+      start_date = raw_query['data']['search']['records'].first['dates'].first['range']['gte']
+      end_date = raw_query['data']['search']['records'].first['dates'].first['range']['lte']
+      normalized = NormalizeTimdex.new.to_result(raw_query, 'Timurid Architecture Research Archive')
+      assert_equal "#{start_date}-#{end_date}", normalized['results'][0].year
+    end
+  end
 end