Revert PR #8723 which disabled the smart JSON chunker when the docume…

…nt contained large strings.

This is now safe.

go/store/prolly/tree/json_chunker.go

@@ -56,13 +56,6 @@ func SerializeJsonToAddr(ctx context.Context, ns NodeStore, j sql.JSONWrapper) (
 	}
 	jsonChunker.appendJsonToBuffer(jsonBytes)
 	err = jsonChunker.processBuffer(ctx)
-	if largeJsonStringError.Is(err) {
-		// Due to current limits on chunk sizes, and an inability of older clients to read
-		// string keys and values split across multiple chunks, we can't use the JSON chunker for documents
-		// with extremely long strings.
-		node, _, err := serializeJsonToBlob(ctx, ns, j)
-		return node, err
-	}
 	if err != nil {
 		return Node{}, err
 	}

go/store/prolly/tree/json_indexed_document.go

@@ -114,7 +114,7 @@ func tryWithFallback(
 	tryFunc func() error,
 	fallbackFunc func(document types.JSONDocument) error) error {
 	err := tryFunc()
-	if err == unknownLocationKeyError || err == unsupportedPathError || err == jsonParseError || largeJsonStringError.Is(err) {
+	if err == unknownLocationKeyError || err == unsupportedPathError || err == jsonParseError {
 		if err != unsupportedPathError {
 			if sqlCtx, ok := ctx.(*sql.Context); ok {
 				sqlCtx.GetLogger().Warn(err)

go/store/prolly/tree/json_scanner.go

@@ -17,8 +17,6 @@ package tree
 import (
 	"fmt"
 	"io"
-
-	errorkinds "gopkg.in/src-d/go-errors.v1"
 )
 
 // JsonScanner is a state machine that parses already-normalized JSON while keeping track of the path to the current value.
@@ -40,15 +38,7 @@ type JsonScanner struct {
 	valueOffset int
 }
 
-// The JSONChunker can't draw a chunk boundary within an object key.
-// This can lead to large chunks that may cause problems.
-// We've observed chunks getting written incorrectly if they exceed 48KB.
-// Since boundaries are always drawn once a chunk exceeds maxChunkSize (16KB),
-// this is the largest length that can be appended to a chunk without exceeding 48KB.
-var maxJsonStringLength = 48*1024 - maxChunkSize
-
 var jsonParseError = fmt.Errorf("encountered invalid JSON while reading JSON from the database, or while preparing to write JSON to the database. This is most likely a bug in JSON diffing")
-var largeJsonStringError = errorkinds.NewKind("encountered JSON key with length %s, larger than max allowed length %s")
 
 func (j JsonScanner) Clone() JsonScanner {
 	return JsonScanner{
@@ -215,9 +205,6 @@ func (s *JsonScanner) acceptKeyString() (stringBytes []byte, err error) {
 		}
 		s.valueOffset++
 	}
-	if s.valueOffset-stringStart > maxJsonStringLength {
-		return nil, largeJsonStringError.New(s.valueOffset-stringStart, maxJsonStringLength)
-	}
 	result := s.jsonBuffer[stringStart:s.valueOffset]
 	// Advance past the ending quotes
 	s.valueOffset++
@@ -233,20 +220,13 @@ func (s *JsonScanner) acceptValueString() (finishedString bool, err error) {
 }
 
 func (s *JsonScanner) acceptRestOfValueString() (finishedString bool, err error) {
-	stringStart := s.valueOffset
 	for s.current() != '"' {
 		switch s.current() {
 		case '\\':
 			s.valueOffset++
 		}
 		s.valueOffset++
 	}
-	// We don't currently split value strings across chunks because it causes issues being read by older clients.
-	// Instead, by returning largeJsonStringError, we trigger the fallback behavior where the JSON document
-	// gets treated as a non-indexed blob.
-	if s.valueOffset-stringStart > maxJsonStringLength {
-		return false, largeJsonStringError.New(s.valueOffset-stringStart, maxJsonStringLength)
-	}
 	// Advance past the ending quotes
 	s.valueOffset++
 	return true, nil

integration-tests/bats/json.bats

@@ -298,24 +298,3 @@ SQL
     [ "$status" -eq 0 ]
     [[ "$output" =~ "Blob" ]] || false
 }
-
-# Older clients have a bug that prevents them from reading indexed JSON documents with strings split across chunks,
-# And making one large chunk can causes issues with the journal.
-# Thus, we expect that the document gets stored as a blob.
-@test "json: document with large string value doesn't get indexed" {
-    run dolt sql <<SQL
-    CREATE TABLE js (
-        pk int PRIMARY KEY,
-        js json
-    );
-    insert into js values (1, '{"key": "`head -c 2097152 < /dev/zero | tr '\0' '\141'`" }');
-    insert into js values (2, '{"`head -c 2097152 < /dev/zero | tr '\0' '\141'`": "value" }');
-SQL
-    # If the documents aren't put into an IndexedJsonDocument, they will contain the following hashes.
-    run dolt show 9erat0qpsqrmmr53fhu6b7gnjj5n1v9i
-    [ "$status" -eq 0 ]
-    [[ "$output" =~ "Blob" ]] || false
-    run dolt show 69cm31n0k392sch3snf5jc7ndfiaund8
-    [ "$status" -eq 0 ]
-    [[ "$output" =~ "Blob" ]] || false
-}