Factor profile days logic into 2 udfs

sql/clients_last_seen_raw_v1.sql

@@ -10,8 +10,23 @@ CREATE TEMP FUNCTION
       0));
   --
 CREATE TEMP FUNCTION
-  udf_combine_days(prev INT64,
+  udf_combine_adjacent_days_bits(prev INT64,
     curr INT64) AS (udf_shift_one_day(prev) + IFNULL(curr,
+    0));
+  --
+CREATE TEMP FUNCTION
+  udf_coalesce_adjacent_days_bits(prev INT64,
+    curr INT64) AS ( COALESCE( NULLIF(udf_shift_one_day(prev),
+        0),
+      curr,
+      0));
+  --
+CREATE TEMP FUNCTION
+  udf_bits_from_days_since_created_profile(days_since_created_profile INT64) AS (
+  IF
+    (days_since_created_profile BETWEEN 0
+      AND 6,
+      1 << days_since_created_profile,
       0));
   --
 CREATE TEMP FUNCTION
@@ -33,7 +48,7 @@ CREATE TEMP FUNCTION
       SELECT
         AS STRUCT experiment,
         branch,
-        udf_combine_days(prev.bits,
+        udf_combine_adjacent_days_bits(prev.bits,
           curr.bits) AS bits
       FROM
         UNNEST(prev) AS prev
@@ -43,7 +58,7 @@ CREATE TEMP FUNCTION
         (experiment,
           branch)
       WHERE
-        udf_combine_days(prev.bits,
+        udf_combine_adjacent_days_bits(prev.bits,
           curr.bits) > 0),
       -- Experiments present in curr only
       ARRAY(
@@ -75,9 +90,9 @@ WITH
     CAST(devtools_toolbox_opened_count_sum > 0 AS INT64) AS days_opened_dev_tools_bits,
     -- We only trust profile_date if it is within one week of the ping submission,
     -- so we ignore any value more than seven days old.
-    IFNULL(1 << DATE_DIFF(submission_date_s3, SAFE.PARSE_DATE("%F",
-          SUBSTR(profile_creation_date, 0, 10)), DAY),
-      0) & udf_bitmask_lowest_7() AS days_created_profile_bits,
+    udf_bits_from_days_since_created_profile(
+      DATE_DIFF(submission_date_s3, SAFE.PARSE_DATE("%F",
+        SUBSTR(profile_creation_date, 0, 10)), DAY)) AS days_created_profile_bits,
     -- Experiments are an array, so we keep track of a usage bit pattern per experiment.
     ARRAY(
     SELECT
@@ -109,18 +124,13 @@ IF
   (_current.client_id IS NOT NULL,
     _current,
     _previous).* REPLACE (
-      udf_combine_days(_previous.days_seen_bits,
+      udf_combine_adjacent_days_bits(_previous.days_seen_bits,
         _current.days_seen_bits) AS days_seen_bits,
-      udf_combine_days(_previous.days_visited_5_uri_bits,
+      udf_combine_adjacent_days_bits(_previous.days_visited_5_uri_bits,
         _current.days_visited_5_uri_bits) AS days_visited_5_uri_bits,
-      udf_combine_days(_previous.days_opened_dev_tools_bits,
+      udf_combine_adjacent_days_bits(_previous.days_opened_dev_tools_bits,
         _current.days_opened_dev_tools_bits) AS days_opened_dev_tools_bits,
-      -- We want to base new profile creation date on the first profile_creation_date
-      -- value we observe, so we propagate an existing non-null value in preference
-      -- to a non-null value on today's observation.
-      COALESCE( --
-        NULLIF(udf_shift_one_day(_previous.days_created_profile_bits),
-          0),
+      udf_coalesce_adjacent_days_bits(_previous.days_created_profile_bits,
         _current.days_created_profile_bits) AS days_created_profile_bits,
       udf_combine_experiment_days(_previous.days_seen_in_experiment,
         _current.days_seen_in_experiment) AS days_seen_in_experiment)

udf/udf_bits_from_days_since_created_profile.sql

@@ -0,0 +1,31 @@
+CREATE TEMP FUNCTION
+  udf_bits_from_days_since_created_profile(days_since_created_profile INT64) AS (
+  IF
+    (days_since_created_profile BETWEEN 0
+      AND 6,
+      1 << days_since_created_profile,
+      0));
+
+/*
+
+Takes in a difference between submission date and profile creation date
+and returns a bit pattern representing the profile creation date IFF
+the profile date is the same as the submission date or no more than
+6 days earlier.
+
+Analysis has shown that client-reported profile creation dates are much
+less reliable outside of this range and cannot be used as reliable indicators
+of new profile creation.
+
+Example:
+
+SELECT
+  udf_bits_from_days_since_created_profile(0),
+  udf_bits_from_days_since_created_profile(1),
+  udf_bits_from_days_since_created_profile(6)
+  udf_bits_from_days_since_created_profile(-1),
+  udf_bits_from_days_since_created_profile(NULL),
+  udf_bits_from_days_since_created_profile(7);
+1, 2, 64, 0, 0, 0
+
+*/

udf/udf_coalesce_days_bits.sql

@@ -0,0 +1,40 @@
+CREATE TEMP FUNCTION
+  udf_bitmask_lowest_28() AS (0x0FFFFFFF);
+  --
+CREATE TEMP FUNCTION
+  udf_shift_one_day(x INT64) AS (IFNULL((x << 1) & udf_bitmask_lowest_28(),
+      0));
+  --
+CREATE TEMP FUNCTION
+  udf_coalesce_adjacent_days_bits(prev INT64,
+    curr INT64) AS ( COALESCE( NULLIF(udf_shift_one_day(prev),
+        0),
+      curr,
+      0));
+
+/*
+
+We generally want to believe only the first reasonable profile creation
+date that we receive from a client.
+Given bits representing usage from the previous day and the current day,
+this function shifts the first argument by one day and returns either that
+value if non-zero and non-null, the current day value if non-zero and non-null,
+or else 0.
+
+Example:
+
+SELECT
+  udf_coalesce_adjacent_days_bits(1,
+    64),
+  udf_coalesce_adjacent_days_bits(64,
+    1),
+  udf_coalesce_adjacent_days_bits(0,
+    64),
+  udf_coalesce_adjacent_days_bits(NULL,
+    64),
+  udf_coalesce_adjacent_days_bits(NULL,
+    NULL);
+2, 128, 64, 64, 0
+
+*/
+