Fix for casting tdigest to text and vice versa

tdigest.c

@@ -1689,16 +1689,15 @@ tdigest_in(PG_FUNCTION_ARGS)
 	tdigest_t  *digest = NULL;
 
 	/* t-digest header fields */
+	int32       flags;
 	int64		count;
 	int			compression;
 	int			ncentroids;
-	char	   *centroids;
+	int			header_length;
 	char	   *ptr;
 
-	centroids = palloc(strlen(str));
-
-	r = sscanf(str, "count %ld compression %d centroids %d%s",
-			   &count, &compression, &ncentroids, centroids);
+	r = sscanf(str, "flags %d count %ld compression %d centroids %d%n",
+			   &flags, &count, &compression, &ncentroids, &header_length);
 
 	if (r != 4)
 		elog(ERROR, "failed to parse t-digest value");
@@ -1725,11 +1724,12 @@ tdigest_in(PG_FUNCTION_ARGS)
 
 	digest = tdigest_allocate(ncentroids);
 
+	digest->flags = flags;
 	digest->count = count;
 	digest->ncentroids = ncentroids;
 	digest->compression = compression;
 
-	ptr = centroids;
+	ptr = str + header_length;
 
 	for (i = 0; i < digest->ncentroids; i++)
 	{
@@ -1750,9 +1750,7 @@ tdigest_in(PG_FUNCTION_ARGS)
 		ptr = strchr(ptr, ')') + 1;
 	}
 
-	Assert(ptr == centroids + strlen(centroids));
-
-	pfree(centroids);
+	Assert(ptr == str + strlen(str));
 
 	AssertCheckTDigest(digest);
 
@@ -1803,7 +1801,7 @@ tdigest_recv(PG_FUNCTION_ARGS)
 	if (flags != 0)
 		elog(ERROR, "unsupported t-digest on-disk format");
 
-	count = pq_getmsgint(buf, sizeof(int64));
+	count = pq_getmsgint64(buf);
 	compression = pq_getmsgint(buf, sizeof(int32));
 	ncentroids = pq_getmsgint(buf, sizeof(int32));
 
@@ -1817,7 +1815,7 @@ tdigest_recv(PG_FUNCTION_ARGS)
 	for (i = 0; i < digest->ncentroids; i++)
 	{
 		digest->centroids[i].sum = pq_getmsgfloat8(buf);
-		digest->centroids[i].count = pq_getmsgint(buf, sizeof(int64));
+		digest->centroids[i].count = pq_getmsgint64(buf);
 	}
 
 	PG_RETURN_POINTER(digest);

test/expected/tdigest.out

@@ -1196,3 +1196,65 @@ SELECT * FROM (
 ---+---+---
 (0 rows)
 
+-- some basic tests to verify transforming from and to text work
+-- 10 centroids (tiny)
+WITH data AS (SELECT i / 1000000.0 AS x FROM generate_series(1,1000000) s(i)),
+     intermediate AS (SELECT tdigest(x, 10)::text AS intermediate_x FROM data),
+     tdigest_parsed AS (SELECT tdigest_percentile(intermediate_x::tdigest, ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS a FROM intermediate),
+     pg_percentile AS (SELECT percentile_cont(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) WITHIN GROUP (ORDER BY x) AS b FROM data)
+SELECT
+    p,
+    abs(a - b) < 0.01, -- arbitrary threshold of 1%
+    (CASE WHEN abs(a - b) < 0.01 THEN NULL ELSE (a - b) END) AS err
+FROM (
+    SELECT
+        unnest(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS p,
+        unnest(a) AS a,
+        unnest(b) AS b
+    FROM tdigest_parsed,
+         pg_percentile
+) foo;
+  p   | ?column? | err 
+------+----------+-----
+ 0.01 | t        |    
+ 0.05 | t        |    
+  0.1 | t        |    
+  0.9 | t        |    
+ 0.95 | t        |    
+ 0.99 | t        |    
+(6 rows)
+
+-- verify we can store tdigest in a summary table
+CREATE TABLE intermediate_tdigest (grouping int, summary tdigest);
+WITH data AS (SELECT row_number() OVER () AS i, pow(z, 4) AS x FROM random_normal(1000000) s(z))
+INSERT INTO intermediate_tdigest
+SELECT
+    i % 10 AS grouping,
+    tdigest(x, 100) AS summary
+FROM data
+GROUP BY i % 10;
+WITH data AS (SELECT pow(z, 4) AS x FROM random_normal(1000000) s(z)),
+     intermediate AS (SELECT tdigest_percentile(summary, ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS a FROM intermediate_tdigest),
+     pg_percentile AS (SELECT percentile_cont(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) WITHIN GROUP (ORDER BY x) AS b FROM data)
+SELECT
+    p,
+    abs(a - b) < 0.01, -- arbitrary threshold of 1%
+    (CASE WHEN abs(a - b) < 0.01 THEN NULL ELSE (a - b) END) AS err
+FROM (
+    SELECT
+        unnest(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS p,
+        unnest(a) AS a,
+        unnest(b) AS b
+    FROM intermediate,
+         pg_percentile
+) foo;
+  p   | ?column? | err 
+------+----------+-----
+ 0.01 | t        |    
+ 0.05 | t        |    
+  0.1 | t        |    
+  0.9 | t        |    
+ 0.95 | t        |    
+ 0.99 | t        |    
+(6 rows)
+

test/sql/tdigest.sql

@@ -892,3 +892,49 @@ SELECT * FROM (
             unnest(tdigest_percentile(x, 1000, (SELECT p FROM perc))) AS a
         FROM data
     ) foo ) bar WHERE a <= b;
+
+-- some basic tests to verify transforming from and to text work
+-- 10 centroids (tiny)
+WITH data AS (SELECT i / 1000000.0 AS x FROM generate_series(1,1000000) s(i)),
+     intermediate AS (SELECT tdigest(x, 10)::text AS intermediate_x FROM data),
+     tdigest_parsed AS (SELECT tdigest_percentile(intermediate_x::tdigest, ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS a FROM intermediate),
+     pg_percentile AS (SELECT percentile_cont(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) WITHIN GROUP (ORDER BY x) AS b FROM data)
+SELECT
+    p,
+    abs(a - b) < 0.01, -- arbitrary threshold of 1%
+    (CASE WHEN abs(a - b) < 0.01 THEN NULL ELSE (a - b) END) AS err
+FROM (
+    SELECT
+        unnest(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS p,
+        unnest(a) AS a,
+        unnest(b) AS b
+    FROM tdigest_parsed,
+         pg_percentile
+) foo;
+
+-- verify we can store tdigest in a summary table
+CREATE TABLE intermediate_tdigest (grouping int, summary tdigest);
+
+WITH data AS (SELECT row_number() OVER () AS i, pow(z, 4) AS x FROM random_normal(1000000) s(z))
+INSERT INTO intermediate_tdigest
+SELECT
+    i % 10 AS grouping,
+    tdigest(x, 100) AS summary
+FROM data
+GROUP BY i % 10;
+
+WITH data AS (SELECT pow(z, 4) AS x FROM random_normal(1000000) s(z)),
+     intermediate AS (SELECT tdigest_percentile(summary, ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS a FROM intermediate_tdigest),
+     pg_percentile AS (SELECT percentile_cont(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) WITHIN GROUP (ORDER BY x) AS b FROM data)
+SELECT
+    p,
+    abs(a - b) < 0.01, -- arbitrary threshold of 1%
+    (CASE WHEN abs(a - b) < 0.01 THEN NULL ELSE (a - b) END) AS err
+FROM (
+    SELECT
+        unnest(ARRAY[0.01, 0.05, 0.1, 0.9, 0.95, 0.99]) AS p,
+        unnest(a) AS a,
+        unnest(b) AS b
+    FROM intermediate,
+         pg_percentile
+) foo;