From fbd54f3857df8b8d31f9ef1675a8933870488888 Mon Sep 17 00:00:00 2001
From: Ahsan Barkati <ahsanbarkati@gmail.com>
Date: Wed, 31 Mar 2021 13:31:38 +0530
Subject: [PATCH] fix(query): Fix pagination with match functions (#7668)

Fix queries involving `regexp`, `allofterms`, `alloftext` and `match`
function with pagination. These functions relies on indexes and
needs to fetch postings/uids for each relevant index key to generate
the final result. Remove early pagination for these cases.
---
 query/common_test.go | 23 ++++++++++++++++++++
 query/query.go       | 27 +++++++++++++----------
 query/query0_test.go | 52 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 90 insertions(+), 12 deletions(-)

diff --git a/query/common_test.go b/query/common_test.go
index 34b93e65823..de2070c8d13 100644
--- a/query/common_test.go
+++ b/query/common_test.go
@@ -328,6 +328,10 @@ gender                         : string .
 indexpred                      : string @index(exact) .
 pred                           : string .
 pname                          : string .
+tweet-a                        : string @index(trigram) .
+tweet-b                        : string @index(term) .
+tweet-c                        : string @index(fulltext) .
+tweet-d                        : string @index(trigram) .
 `
 
 func populateCluster() {
@@ -828,6 +832,25 @@ func populateCluster() {
 		<67> <index-pred2> "I" .
 		<68> <index-pred2> "J" .
 		<69> <index-pred2> "K" .
+
+		<61> <tweet-a> "aaa" .
+		<62> <tweet-a> "aaaa" .
+		<63> <tweet-a> "aaaab" .
+		<64> <tweet-a> "aaaabb" .
+
+		<61> <tweet-b> "indiana" .
+		<62> <tweet-b> "indiana" .
+		<63> <tweet-b> "indiana jones" .
+		<64> <tweet-b> "indiana pop" .
+
+		<61> <tweet-c> "I am a citizen" .
+		<62> <tweet-c> "I am a citizen" .
+		<63> <tweet-c> "I am a citizen" .
+		<64> <tweet-c> "I am a citizen of Paradis Island" .
+
+		<61> <tweet-d> "aaabxxx" .
+		<62> <tweet-d> "aaacdxx" .
+		<63> <tweet-d> "aaabcd" .
 	`)
 	if err != nil {
 		panic(fmt.Sprintf("Could not able add triple to the cluster. Got error %v", err.Error()))
diff --git a/query/query.go b/query/query.go
index 9368da565a6..828ae60c8b1 100644
--- a/query/query.go
+++ b/query/query.go
@@ -984,19 +984,22 @@ func calculateFirstN(sg *SubGraph) int32 {
 	//     name
 	//   }
 	// }
-	// - should be has function (Right now, I'm doing it for has, later it can be extended)
-	// {
-	//   q(func: has(name), first:1) {
-	//     name
-	//   }
-	// }
-	// isSupportedFunction := sg.SrcFunc != nil && sg.SrcFunc.Name == "has"
+	// - should not be one of those function which fetches some results and then do further
+	// processing to narrow down the result. For example: allofterm will fetch the index postings
+	// for each term and then do an intersection.
+	// TODO: Look into how we can optimize queries involving these functions.
+
+	shouldExclude := false
+	if sg.SrcFunc != nil {
+		switch sg.SrcFunc.Name {
+		case "regexp", "alloftext", "allofterms", "match":
+			shouldExclude = true
+		default:
+			shouldExclude = false
+		}
+	}
 
-	// Manish: Shouldn't all functions allow this? If we don't have a order and we don't have a
-	// filter, then we can respect the first N, offset Y arguments when retrieving data.
-	isSupportedFunction := true
-	if len(sg.Filters) == 0 && len(sg.Params.Order) == 0 &&
-		isSupportedFunction {
+	if len(sg.Filters) == 0 && len(sg.Params.Order) == 0 && !shouldExclude {
 		// Offset also added because, we need n results to trim the offset.
 		if sg.Params.Count != 0 {
 			count = sg.Params.Count + sg.Params.Offset
diff --git a/query/query0_test.go b/query/query0_test.go
index cc52c04b843..118e0f938a1 100644
--- a/query/query0_test.go
+++ b/query/query0_test.go
@@ -3432,6 +3432,58 @@ func TestEqFilterWithoutIndex(t *testing.T) {
 
 }
 
+func TestMatchingWithPagination(t *testing.T) {
+	tests := []struct {
+		name     string
+		query    string
+		expected string
+	}{
+		{
+			`Test regexp matching with pagination`,
+			`{
+				me(func: regexp(tweet-a, /aaa.b/), first:1){
+					tweet-a
+				}
+			 }`,
+			`{"data":{"me":[{"tweet-a":"aaaab"}]}}`,
+		},
+		{
+			`Test term matching with pagination`,
+			`{
+				me(func: allofterms(tweet-b, "indiana jones"), first:1){
+					tweet-b
+				}
+			 }`,
+			`{"data":{"me":[{"tweet-b":"indiana jones"}]}}`,
+		},
+		{
+			`Test full-text matching with pagination`,
+			`{
+				me(func: alloftext(tweet-c, "I am a citizen of Paradis Island"), first:1){
+					tweet-c
+				}
+			 }`,
+			`{"data":{"me":[{"tweet-c":"I am a citizen of Paradis Island"}]}}`,
+		},
+		{
+			`Test match function with pagination`,
+			`{
+				me(func: match(tweet-d, "aaaaaa", 3), first:1) {
+					tweet-d
+				}
+			 }`,
+			`{"data":{"me":[{"tweet-d":"aaabcd"}]}}`,
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			result := processQueryNoErr(t, tc.query)
+			require.JSONEq(t, tc.expected, result)
+		})
+	}
+}
+
 var client *dgo.Dgraph
 
 func TestMain(m *testing.M) {