colblk: use configured Compare in index blocks

Previously columnar index blocks always used bytes.Compare during seeks among
separators.

sstable/colblk/index_block.go

@@ -5,8 +5,6 @@
 package colblk
 
 import (
-	"bytes"
-
 	"github.com/cockroachdb/errors"
 	"github.com/cockroachdb/pebble/internal/base"
 	"github.com/cockroachdb/pebble/internal/binfmt"
@@ -178,9 +176,10 @@ func (r *IndexReader) Describe(f *binfmt.Formatter) {
 
 // IndexIter is an iterator over the block entries in an index block.
 type IndexIter struct {
-	r   *IndexReader
-	n   int
-	row int
+	compare base.Compare
+	r       *IndexReader
+	n       int
+	row     int
 
 	h           block.BufferHandle
 	allocReader IndexReader
@@ -190,8 +189,14 @@ type IndexIter struct {
 var _ block.IndexBlockIterator = (*IndexIter)(nil)
 
 // InitReader initializes an index iterator from the provided reader.
-func (i *IndexIter) InitReader(r *IndexReader) {
-	*i = IndexIter{r: r, n: int(r.br.header.Rows), h: i.h, allocReader: i.allocReader}
+func (i *IndexIter) InitReader(compare base.Compare, r *IndexReader) {
+	*i = IndexIter{
+		compare:     compare,
+		r:           r,
+		n:           int(r.br.header.Rows),
+		h:           i.h,
+		allocReader: i.allocReader,
+	}
 }
 
 // Init initializes an iterator from the provided block data slice.
@@ -202,7 +207,7 @@ func (i *IndexIter) Init(
 	i.h = block.BufferHandle{}
 	// TODO(jackson): Handle the transforms.
 	i.allocReader.Init(blk)
-	i.InitReader(&i.allocReader)
+	i.InitReader(cmp, &i.allocReader)
 	return nil
 }
 
@@ -220,7 +225,7 @@ func (i *IndexIter) InitHandle(
 	i.h.Release()
 	i.h = blk
 	i.allocReader.Init(i.h.Get())
-	i.InitReader(&i.allocReader)
+	i.InitReader(cmp, &i.allocReader)
 	return nil
 }
 
@@ -295,7 +300,7 @@ func (i *IndexIter) SeekGE(key []byte) bool {
 
 		// TODO(jackson): Is Bytes.At or Bytes.Slice(Bytes.Offset(h),
 		// Bytes.Offset(h+1)) faster in this code?
-		c := bytes.Compare(key, i.r.separators.At(h))
+		c := i.compare(key, i.r.separators.At(h))
 		if c > 0 {
 			index = h + 1 // preserves f(index-1) == false
 		} else {

sstable/colblk/index_block_test.go

@@ -53,7 +53,7 @@ func TestIndexBlock(t *testing.T) {
 			return buf.String()
 		case "iter":
 			var it IndexIter
-			it.InitReader(&r)
+			it.InitReader(testkeys.Comparer.Compare, &r)
 			for _, line := range strings.Split(d.Input, "\n") {
 				fields := strings.Fields(line)
 				var valid bool

sstable/colblk/testdata/index_block

@@ -419,3 +419,79 @@ UnsafeSeparator(3) = "cephalopod"
 103-103: x                      # data[2]:
 103-106: x 627034               # data[3]: bp4
 106-107: x 00                   # block padding byte
+
+build
+cat@20 3021   2052   bp1
+cat@10 91251   1899
+cat@5 91251   1899
+----
+UnsafeSeparator(2) = "cat@5"
+# index block header
+# columnar block header
+00-01: x 01           # version 1
+01-03: x 0400         # 4 columns
+03-07: x 03000000     # 3 rows
+07-08: b 00000011     # col 0: bytes
+08-12: x 1b000000     # col 0: page start 27
+12-13: b 00000010     # col 1: uint
+13-17: x 31000000     # col 1: page start 49
+17-18: b 00000010     # col 2: uint
+18-22: x 40000000     # col 2: page start 64
+22-23: b 00000011     # col 3: bytes
+23-27: x 48000000     # col 3: page start 72
+# data for column 0
+# rawbytes
+# offsets table
+27-28: x 01           # encoding: 1b
+28-29: x 00           # data[0] = 0 [32 overall]
+29-30: x 06           # data[1] = 6 [38 overall]
+30-31: x 0c           # data[2] = 12 [44 overall]
+31-32: x 11           # data[3] = 17 [49 overall]
+# data
+32-38: x 636174403230 # data[0]: cat@20
+38-44: x 636174403130 # data[1]: cat@10
+44-49: x 6361744035   # data[2]: cat@5
+# data for column 1
+49-50: x 04           # encoding: 4b
+50-52: x 0000         # padding (aligning to 32-bit boundary)
+52-56: x cd0b0000     # data[0] = 3021
+56-60: x 73640100     # data[1] = 91251
+60-64: x 73640100     # data[2] = 91251
+# data for column 2
+64-65: x 02           # encoding: 2b
+65-66: x 00           # padding (aligning to 16-bit boundary)
+66-68: x 0408         # data[0] = 2052
+68-70: x 6b07         # data[1] = 1899
+70-72: x 6b07         # data[2] = 1899
+# data for column 3
+# rawbytes
+# offsets table
+72-73: x 01           # encoding: 1b
+73-74: x 00           # data[0] = 0 [77 overall]
+74-75: x 03           # data[1] = 3 [80 overall]
+75-76: x 03           # data[2] = 3 [80 overall]
+76-77: x 03           # data[3] = 3 [80 overall]
+# data
+77-80: x 627031       # data[0]: bp1
+80-80: x              # data[1]:
+80-80: x              # data[2]:
+80-81: x 00           # block padding byte
+
+iter
+seek-ge cat
+seek-ge cat@21
+seek-ge cat@20
+seek-ge cat@19
+seek-ge cat@10
+seek-ge cat@9
+seek-ge cat@5
+seek-ge cat@2
+----
+block 0: 3021-5073 props="bp1"
+block 0: 3021-5073 props="bp1"
+block 0: 3021-5073 props="bp1"
+block 1: 91251-93150
+block 1: 91251-93150
+block 2: 91251-93150
+block 2: 91251-93150
+.