chore(lsm): cleanup and add tools

.gitignore

@@ -1,3 +1,4 @@
 test.db
 *.mercury
 sour.is-mercury
+.vscode/

lsm/cli/main.go

@@ -0,0 +1,286 @@
+package main
+
+import (
+	"bytes"
+	"context"
+	"encoding/base64"
+	"errors"
+	"fmt"
+	"io"
+	"iter"
+	"net/http"
+	"net/url"
+	"os"
+	"time"
+
+	"github.com/docopt/docopt-go"
+	"go.sour.is/pkg/lsm"
+)
+
+var usage = `
+Usage: 
+  lsm create <archive> <files>...
+  lsm append <archive> <files>...
+  lsm read <archive> [<start> [<end>]]
+  lsm serve <archive>
+  lsm client <archive> [<start> [<end>]]`
+
+type args struct {
+	Create bool
+	Append bool
+	Read   bool
+	Serve  bool
+	Client bool
+
+	Archive string   `docopt:"<archive>"`
+	Files   []string `docopt:"<files>"`
+	Start   int64    `docopt:"<start>"`
+	End     int64    `docopt:"<end>"`
+}
+
+func main() {
+	opts, err := docopt.ParseDoc(usage)
+	if err != nil {
+		fmt.Println(err)
+		os.Exit(1)
+	}
+	args := args{}
+	err = opts.Bind(&args)
+	if err != nil {
+		fmt.Println(err)
+		os.Exit(1)
+	}
+	err = run(Console, args)
+	if err != nil {
+		fmt.Println(err)
+		os.Exit(1)
+	}
+}
+
+type console struct {
+	Stdin  io.Reader
+	Stdout io.Writer
+	Stderr io.Writer
+}
+var Console = console{os.Stdin, os.Stdout, os.Stderr}
+
+func (c console) Write(b []byte) (int, error) {
+	return c.Stdout.Write(b)
+}
+
+func run(console console, a args) error {
+	fmt.Fprintln(console, "lsm")
+	switch {
+	case a.Create:
+		f, err := os.OpenFile(a.Archive, os.O_CREATE|os.O_WRONLY, 0644)
+		if err != nil {
+			return err
+		}
+		defer f.Close()
+
+		return lsm.WriteLogFile(f, fileReaders(a.Files))
+	case a.Append:
+		f, err := os.OpenFile(a.Archive, os.O_RDWR, 0644)
+		if err != nil {
+			return err
+		}
+		defer f.Close()
+
+		return lsm.AppendLogFile(f, fileReaders(a.Files))
+	case a.Read:
+		fmt.Fprintln(console, "reading", a.Archive)
+
+		f, err := os.Open(a.Archive)
+		if err != nil {
+			return err
+		}
+		defer f.Close()
+		return readContent(f, console, a.Start, a.End)
+	case a.Serve:
+		fmt.Fprintln(console, "serving", a.Archive)
+		b, err := base64.RawStdEncoding.DecodeString(a.Archive)
+		now := time.Now()
+		if err != nil {
+			return err
+		}
+		http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
+			http.ServeContent(w, r, "", now, bytes.NewReader(b))
+		})
+		return http.ListenAndServe(":8080", nil)
+	case a.Client:
+		r, err := OpenHttpReader(context.Background(), a.Archive, 0)
+		if err != nil {
+			return err
+		}
+		defer r.Close()
+		defer func() {fmt.Println("bytes read", r.bytesRead)}()
+		return readContent(r, console, a.Start, a.End)
+	}
+	return errors.New("unknown command")
+}
+
+func readContent(r io.ReaderAt, console console, start, end int64) error {
+	lg, err := lsm.ReadLogFile(r)
+	if err != nil {
+		return err
+	}
+	for bi, rd := range lg.Iter(uint64(start)) {
+		if end > 0 && int64(bi.Index) >= end {
+			break
+		}
+		fmt.Fprintf(console, "=========================\n%+v:\n", bi)
+		wr := base64.NewEncoder(base64.RawStdEncoding, console)
+		io.Copy(wr, rd)
+		fmt.Fprintln(console, "\n=========================")
+	}
+	if lg.Err != nil {
+		return lg.Err
+	}
+	for bi, rd := range lg.Rev(lg.Count()) {
+		if end > 0 && int64(bi.Index) >= end {
+			break
+		}
+		fmt.Fprintf(console, "=========================\n%+v:\n", bi)
+		wr := base64.NewEncoder(base64.RawStdEncoding, console)
+		io.Copy(wr, rd)
+		fmt.Fprintln(console, "\n=========================")
+	}
+
+	return lg.Err
+}
+
+
+func fileReaders(names []string) iter.Seq[io.Reader] {
+	return iter.Seq[io.Reader](func(yield func(io.Reader) bool) {
+		for _, name := range names {
+			f, err := os.Open(name)
+			if err != nil {
+				continue
+			}
+			if !yield(f) {
+				f.Close()
+				return
+			}
+			f.Close()
+		}
+	})
+}
+
+type HttpReader struct {
+	ctx     context.Context
+	uri     url.URL
+	tmpfile *os.File
+	pos     int64
+	end     int64
+	bytesRead int
+}
+
+func OpenHttpReader(ctx context.Context, uri string, end int64) (*HttpReader, error) {
+	u, err := url.Parse(uri)
+	if err != nil {
+		return nil, err
+	}
+	return &HttpReader{ctx: ctx, uri: *u, end: end}, nil
+}
+
+func (r *HttpReader) Read(p []byte) (int, error) {
+	n, err := r.ReadAt(p, r.pos)
+	if err != nil {
+		return n, err
+	}
+	r.pos += int64(n)
+	r.bytesRead += n
+	return n, nil
+}
+
+func (r *HttpReader) Seek(offset int64, whence int) (int64, error) {
+	switch whence {
+	case io.SeekStart:
+		r.pos = offset
+	case io.SeekCurrent:
+		r.pos += offset
+	case io.SeekEnd:
+		r.pos = r.end + offset
+	}
+	return r.pos, nil
+}
+
+func (r *HttpReader) Close() error {
+	r.ctx.Done()
+	return nil
+}
+
+// ReadAt implements io.ReaderAt. It reads data from the internal buffer starting
+// from the specified offset and writes it into the provided data slice. If the
+// offset is negative, it returns an error. If the requested read extends beyond
+// the buffer's length, it returns the data read so far along with an io.EOF error.
+func (r *HttpReader) ReadAt(data []byte, offset int64) (int, error) {
+	if err := r.ctx.Err(); err != nil {
+		return 0, err
+	}
+
+	if offset < 0 {
+		return 0, errors.New("negative offset")
+	}
+
+	if r.end > 0 && offset > r.end {
+		return 0, io.EOF
+	}
+
+	dlen := len(data) + int(offset)
+
+	if r.end > 0 && r.end+int64(dlen) > r.end {
+		dlen = int(r.end)
+	}
+
+	end := ""
+	if r.end > 0 {
+		end = fmt.Sprintf("/%d", r.end)
+	}
+	req, err := http.NewRequestWithContext(r.ctx, "GET", r.uri.String(), nil)
+	if err != nil {
+		return 0, err
+	}
+
+	req.Header.Set("Range", fmt.Sprintf("bytes=%d-%d%s", offset, dlen, end))
+
+	fmt.Fprintln(Console.Stderr, req)
+
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return 0, err
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode == http.StatusRequestedRangeNotSatisfiable {
+		fmt.Fprintln(Console.Stderr, "requested range not satisfiable")
+		return 0, io.EOF
+	}
+
+	if resp.StatusCode == http.StatusOK {
+
+		r.tmpfile, err = os.CreateTemp("", "httpReader")
+		if err != nil {
+			return 0, err
+		}
+		defer os.Remove(r.tmpfile.Name())
+		n, err := io.Copy(r.tmpfile, resp.Body)
+		if err != nil {
+			return 0, err
+		}
+		r.bytesRead += int(n)
+
+		defer fmt.Fprintln(Console.Stderr, "wrote ", n, " bytes to ", r.tmpfile.Name())
+		resp.Body.Close()
+		r.tmpfile.Seek(offset, 0)
+		return io.ReadFull(r.tmpfile, data)
+	}
+
+	n, err := io.ReadFull(resp.Body, data)
+	if n == 0 && err != nil {
+		return n, err
+	}
+	r.bytesRead += n
+	defer fmt.Fprintln(Console.Stderr, "read ", n, " bytes")
+	return n, nil
+}

lsm/marshal.go

@@ -1,138 +0,0 @@
-package lsm
-
-import (
-	"bytes"
-	"encoding"
-	"encoding/binary"
-	"fmt"
-)
-
-type entry struct {
-	key   string
-	value uint64
-}
-
-// MarshalBinary implements encoding.BinaryMarshaler.
-func (e *entry) MarshalBinary() (data []byte, err error) {
-	data = make([]byte, len(e.key), len(e.key)+binary.MaxVarintLen16)
-	copy(data, e.key)
-
-	data = binary.AppendUvarint(data, e.value)
-	reverse(data[len(e.key):])
-	return data, err
-}
-
-// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-func (e *entry) UnmarshalBinary(data []byte) error {
-	// fmt.Println("unmarshal", data, string(data))
-
-	if len(data) < binary.MaxVarintLen16 {
-		return fmt.Errorf("%w: bad data", ErrDecode)
-	}
-	head := make([]byte, binary.MaxVarintLen16)
-	copy(head, data[max(0, len(data)-cap(head)):])
-	reverse(head)
-
-	size := 0
-	e.value, size = binary.Uvarint(head)
-	if size == 0 {
-		return fmt.Errorf("%w: invalid data", ErrDecode)
-	}
-	e.key = string(data[:len(data)-size])
-
-	return nil
-}
-
-var _ encoding.BinaryMarshaler = (*entry)(nil)
-var _ encoding.BinaryUnmarshaler = (*entry)(nil)
-
-type entries []entry
-
-// MarshalBinary implements encoding.BinaryMarshaler.
-func (lis *entries) MarshalBinary() (data []byte, err error) {
-	var buf bytes.Buffer
-
-	for _, e := range *lis {
-		d, err := e.MarshalBinary()
-		if err != nil {
-			return nil, err
-		}
-
-		_, err = buf.Write(d)
-		if err != nil {
-			return nil, err
-		}
-
-		_, err = buf.Write(reverse(binary.AppendUvarint(make([]byte, 0, binary.MaxVarintLen32), uint64(len(d)))))
-		if err != nil {
-			return nil, err
-		}
-	}
-
-	return buf.Bytes(), err
-}
-
-// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-func (lis *entries) UnmarshalBinary(data []byte) error {
-	head := make([]byte, binary.MaxVarintLen16)
-	pos := uint64(len(data))
-
-	for pos > 0 {
-		copy(head, data[max(0, pos-uint64(cap(head))):])
-		length, size := binary.Uvarint(reverse(head))
-
-		e := entry{}
-		if err := e.UnmarshalBinary(data[max(0, pos-(length+uint64(size))) : pos-uint64(size)]); err != nil {
-			return err
-		}
-		*lis = append(*lis, e)
-
-		pos -= length + uint64(size)
-	}
-	reverse(*lis)
-	return nil
-}
-
-var _ encoding.BinaryMarshaler = (*entries)(nil)
-var _ encoding.BinaryUnmarshaler = (*entries)(nil)
-
-type segment struct {
-	entries entries
-}
-
-// MarshalBinary implements encoding.BinaryMarshaler.
-func (s *segment) MarshalBinary() (data []byte, err error) {
-	head := header{
-		entries: uint64(len(s.entries)),
-	}
-
-	data, err = s.entries.MarshalBinary()
-	if err != nil {
-		return nil, err
-	}
-
-	head.datalen = uint64(len(data))
-
-	h := hash()
-	h.Write(data)
-	head.sig = h.Sum(nil)
-
-	return head.Append(data), err
-}
-
-// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-func (s *segment) UnmarshalBinary(data []byte) error {
-	head, err := ReadHead(data)
-	if err != nil {
-		return err
-	}
-
-	h := hash()
-	h.Write(data[:head.datalen])
-	if !bytes.Equal(head.sig, h.Sum(nil)) {
-		return fmt.Errorf("%w: invalid checksum", ErrDecode)
-	}
-
-	s.entries = make(entries, 0, head.entries)
-	return s.entries.UnmarshalBinary(data[:head.datalen])
-}

lsm/marshal_test.go

@@ -1,76 +0,0 @@
-package lsm
-
-import (
-	"io/fs"
-	"testing"
-
-	"github.com/matryer/is"
-)
-
-func TestEncoding(t *testing.T) {
-	is := is.New(t)
-
-	data := segment{entries: entries{
-		{"key-1", 1},
-		{"key-2", 2},
-		{"key-3", 3},
-		{"longerkey-4", 65535},
-	}}
-
-	b, err := data.MarshalBinary()
-	is.NoErr(err)
-
-	var got segment
-	err = got.UnmarshalBinary(b)
-	is.NoErr(err)
-
-	is.Equal(data, got)
-}
-
-func TestReverse(t *testing.T) {
-	is := is.New(t)
-
-	got := []byte("gnirts a si siht")
-	reverse(got)
-
-	is.Equal(got, []byte("this is a string"))
-
-	got = []byte("!gnirts a si siht")
-	reverse(got)
-
-	is.Equal(got, []byte("this is a string!"))
-}
-
-func TestFile(t *testing.T) {
-	is := is.New(t)
-
-	entries := entries {
-		{"key-1", 1},
-		{"key-2", 2},
-		{"key-3", 3},
-		{"longerkey-4", 65535},
-	}
-
-	f := basicFile(t, entries, entries, entries)
-
-	sf, err := ReadFile(f)
-	is.NoErr(err)
-
-	is.Equal(len(sf.segments), 3)
-}
-
-func basicFile(t *testing.T, lis ...entries) fs.File {
-	t.Helper()
-
-	segments := make([][]byte, len(lis))
-	var err error
-	for i, entries := range lis {
-		data := segment{entries: entries}
-		segments[i], err = data.MarshalBinary()
-		if err != nil {
-			t.Error(err)
-		}
-	}
-
-	return NewFile(segments...)
-}

lsm/sst.go

@@ -1,370 +1,634 @@
-// SPDX-FileCopyrightText: 2023 Jon Lundy <jon@xuu.cc>
-// SPDX-License-Identifier: BSD-3-Clause
-
-// lsm -- Log Structured Merge-Tree
-//
-// This is a basic LSM tree using a SSTable optimized for append only writing. On disk data is organized into time ordered
-// files of segments, containing reverse sorted keys. Each segment ends with a magic value `Souris\x01`, a 4byte hash, count of
-// segment entries, and data length.
-
 package lsm
 
 import (
-	"bytes"
-	"encoding"
 	"encoding/binary"
 	"errors"
 	"fmt"
 	"hash/fnv"
 	"io"
-	"io/fs"
+	"iter"
-	"sort"
+	"slices"
+)
+
+// [Sour.is|size] [size|hash][data][hash|flag|size]... [prev|count|flag|size]
+
+// Commit1: [magic>|<end]{10} ... [<count][<size][<flag]{3..30}
+//                 +---------|--------------------------------> end = seek to end of file
+//                       <---|-------------+                    size = seek to magic header
+//                       <---|-------------+10                  size + 10 = seek to start of file
+//          <-----------------------------T+10----------------> 10 + size + trailer = full file size
+
+// Commit2: [magic>|<end]{10} ... [<count][<size][<flag]{3..30} ... [<prev][<count][<size][<flag]{4..40}
+//                           <---|---------+
+//                           <-------------+T----------------->
+//                  +--------|------------------------------------------------------------------------->
+//                           <-------------------------------------|----------------+
+//     prev = seek to last commit                              <---|-+
+//     prev + trailer = size of commit                         <----T+--------------------------------->
+
+// Block:  [hash>|<end]{10} ... [<size][<flag]{2..20}
+//               +---------|------------------------>  end = seek to end of block
+//                         <---|-+                     size = seek to end of header
+//         <-------------------|-+10                   size + 10 = seek to start of block
+//         <---------------------T+10--------------->  size + 10 + trailer = full block size
+
+const (
+	TypeUnknown uint64 = iota
+	TypeSegment
+	TypeCommit
+	TypePrevCommit
+
+	headerSize = 10
+
+	maxCommitSize = 4 * binary.MaxVarintLen64
+	minCommitSize = 3
+
+	maxBlockSize = 2 * binary.MaxVarintLen64
+	minBlockSize = 2
 )
 
 var (
-	magic      = reverse(append([]byte("Souris"), '\x01'))
+	Magic   = [10]byte([]byte("Sour.is\x00\x00\x00"))
-	hash       = fnv.New32a
+	Version = uint8(1)
-	hashLength = hash().Size()
+	hash    = fnv.New64a
-	// segmentSize         = 2 ^ 16 // min 2^9 = 512b, max? 2^20 = 1M
+
-	segmentFooterLength = len(magic) + hashLength + binary.MaxVarintLen32 + binary.MaxVarintLen32
+	ErrDecode = errors.New("decode")
 )
 
 type header struct {
-	sig     []byte // 4Byte signature
+	end   uint64
-	entries uint64 // count of entries in segment
+	extra []byte
-	datalen uint64 // length of data
-	headlen uint64 // length of header
-	end     int64  // location of end of data/start of header (start of data is `end - datalen`)
 }
 
-// ReadHead parse header from a segment. reads from the end of slice of length segmentFooterLength
+// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-func ReadHead(data []byte) (*header, error) {
+// It decodes the input binary data into the header struct.
-	if len(data) < len(magic)+6 {
+// The function expects the input data to be of a specific size (headerSize),
-		return nil, fmt.Errorf("%w: invalid size", ErrDecode)
+// otherwise it returns an error indicating bad data.
+// It reads the 'end' field from the binary data, updates the 'extra' field,
+// and reverses the byte order of 'extra' in place.
+func (h *header) UnmarshalBinary(data []byte) error {
+	if len(data) != headerSize {
+		return fmt.Errorf("%w: bad data", ErrDecode)
 	}
 
-	if !bytes.Equal(data[len(data)-len(magic):], magic) {
+	h.extra = make([]byte, headerSize)
-		return nil, fmt.Errorf("%w: invalid header", ErrDecode)
+	copy(h.extra, data)
+
+	var bytesRead int
+	h.end, bytesRead = binary.Uvarint(h.extra)
+	reverse(h.extra)
+	h.extra = h.extra[:min(8,headerSize-bytesRead)]
+
+	return nil
 }
 
-	head := make([]byte, 0, segmentFooterLength)
+type Commit struct {
-	head = reverse(append(head, data[max(0, len(data)-cap(head)-1):]...))
+	flag  uint64 // flag values
-	size, s := binary.Uvarint(head[len(magic)+4:])
+	size  uint64 // size of the trailer
-	length, i := binary.Uvarint(head[len(magic)+4+s:])
+	count uint64 // number of entries
+	prev  uint64 // previous commit
 
-	return &header{
+	tsize int
-		sig:     head[len(magic) : len(magic)+4],
-		entries: size,
-		datalen: length,
-		headlen: uint64(len(magic) + hashLength + s + i),
-		end:     int64(len(data)),
-	}, nil
-}
-func (h *header) Append(data []byte) []byte {
-
-	length := len(data)
-	data = append(data, h.sig...)
-	data = binary.AppendUvarint(data, h.entries)
-	data = binary.AppendUvarint(data, h.datalen)
-	reverse(data[length:])
-
-	return append(data, magic...)
 }
 
-var _ encoding.BinaryMarshaler = (*segment)(nil)
+// Append marshals the trailer into binary form and appends it to data.
-var _ encoding.BinaryUnmarshaler = (*segment)(nil)
+// It returns the new slice.
-
+func (h *Commit) AppendTrailer(data []byte) []byte {
-var ErrDecode = errors.New("decode")
+	h.flag |= TypeCommit
-
+	// if h.prev > 0 {
-func reverse[T any](b []T) []T {
+	// 	h.flag |= TypePrevCommit
-	l := len(b)
-	for i := 0; i < l/2; i++ {
-		b[i], b[l-i-1] = b[l-i-1], b[i]
-	}
-	return b
-}
-
-// func clone[T ~[]E, E any](e []E) []E {
-// 	return append(e[0:0:0], e...)
 	// }
 
-type entryBytes []byte
+	size := len(data)
-
+	data = binary.AppendUvarint(data, h.size)
-// KeyValue returns the parsed key and value from an entry
+	data = binary.AppendUvarint(data, h.flag)
-func (e entryBytes) KeyValue() ([]byte, uint64) {
+	data = binary.AppendUvarint(data, h.count)
-	if len(e) < 2 {
+	// if h.prev > 0 {
-		return nil, 0
+	// 	data = binary.AppendUvarint(data, h.prev)
-	}
+	// }
-	head := reverse(append(e[0:0:0], e[max(0, len(e)-binary.MaxVarintLen64):]...))
+	reverse(data[size:])
-	value, i := binary.Uvarint(head)
-	return append(e[0:0:0], e[:len(e)-i]...), value
-}
-
-// NewKeyValue packed into an entry
-func NewKeyValue(key []byte, val uint64) entryBytes {
-	length := len(key)
-	data := append(key[0:0:0], key...)
-	data = binary.AppendUvarint(data, val)
-	reverse(data[length:])
 
 	return data
 }
 
-type listEntries []entryBytes
+// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-
+// It reads a trailer from binary data, and sets the fields
-// WriteTo implements io.WriterTo.
+// of the receiver to the values found in the header.
-func (lis *listEntries) WriteTo(wr io.Writer) (int64, error) {
+func (h *Commit) UnmarshalBinary(data []byte) error {
-	if lis == nil {
+	if len(data) < minCommitSize {
-		return 0, nil
+		return fmt.Errorf("%w: bad data", ErrDecode)
 	}
 
-	head := header{
+	var n int
-		entries: uint64(len(*lis)),
+	h.size, n = binary.Uvarint(data)
-	}
+	data = data[n:]
-	h := hash()
+	h.tsize += n
 
-	wr = io.MultiWriter(wr, h)
+	h.flag, n = binary.Uvarint(data)
+	data = data[n:]
+	h.tsize += n
 
-	var i int64
+	h.count, n = binary.Uvarint(data)
-	for _, b := range *lis {
+	data = data[n:]
-		j, err := wr.Write(b)
+	h.tsize += n
-		i += int64(j)
+
-		if err != nil {
+	// h.prev = h.size
-			return i, err
+	if h.flag&TypePrevCommit == TypePrevCommit {
+		h.prev, n = binary.Uvarint(data)
+		h.tsize += n
 	}
 
-		j, err = wr.Write(reverse(binary.AppendUvarint(make([]byte, 0, binary.MaxVarintLen32), uint64(len(b)))))
+	return nil
-		i += int64(j)
-		if err != nil {
-			return i, err
-		}
-	}
-	head.datalen = uint64(i)
-	head.sig = h.Sum(nil)
-
-	b := head.Append([]byte{})
-	j, err := wr.Write(b)
-	i += int64(j)
-
-	return i, err
 }
 
-var _ sort.Interface = listEntries{}
+type Block struct {
+	header
 
-// Len implements sort.Interface.
+	size uint64
-func (lis listEntries) Len() int {
+	flag uint64
-	return len(lis)
+
+	tsize int
 }
 
-// Less implements sort.Interface.
+func (h *Block) AppendHeader(data []byte) []byte {
-func (lis listEntries) Less(i int, j int) bool {
+	size := len(data)
-	iname, _ := lis[i].KeyValue()
+	data = append(data, make([]byte, 10)...)
-	jname, _ := lis[j].KeyValue()
+	copy(data, h.extra)
+	if h.size == 0 {
+		return data
+	}
+	hdata := binary.AppendUvarint(make([]byte, 0, 10), h.end)
+	reverse(hdata)
+	copy(data[size+10-len(hdata):], hdata)
 
-	return bytes.Compare(iname, jname) < 0
+	return data
 }
 
-// Swap implements sort.Interface.
+// AppendTrailer marshals the footer into binary form and appends it to data.
-func (lis listEntries) Swap(i int, j int) {
+// It returns the new slice.
-	lis[i], lis[j] = lis[j], lis[i]
+func (h *Block) AppendTrailer(data []byte) []byte {
+	size := len(data)
+
+	h.flag |= TypeSegment
+	data = binary.AppendUvarint(data, h.size)
+	data = binary.AppendUvarint(data, h.flag)
+	reverse(data[size:])
+
+	return data
 }
 
-type segmentReader struct {
+// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-	head *header
+// It reads a footer from binary data, and sets the fields
-	rd   io.ReaderAt
+// of the receiver to the values found in the footer.
+func (h *Block) UnmarshalBinary(data []byte) error {
+	if len(data) < minBlockSize {
+		return fmt.Errorf("%w: bad data", ErrDecode)
 	}
 
-// FirstEntry parses the first segment entry from the end of the segment
+	var n int
-func (s *segmentReader) FirstEntry() (*entryBytes, error) {
+	h.size, n = binary.Uvarint(data)
-	e, _, err := s.readEntryAt(-1)
+	data = data[n:]
-	return e, err
+	h.tsize += n
-}
 
-func (s *segmentReader) VerifyHash() (bool, error) {
+	h.flag, n = binary.Uvarint(data)
-	h := hash()
+	h.tsize += n
-	data := make([]byte, s.head.datalen)
-	_, err := s.rd.ReadAt(data, s.head.end-int64(s.head.datalen))
-	if err != nil {
-		return false, err
-	}
-	_, err = h.Write(data)
-	ok := bytes.Equal(h.Sum(nil), s.head.sig)
 
-	return ok, err
+	return nil
-}
-
-// Find locates needle within a segment. if it cant find it will return the nearest key before needle.
-func (s *segmentReader) Find(needle []byte, first bool) (*entryBytes, bool, error) {
-	if s == nil {
-		return nil, false, nil
-	}
-	e, pos, err := s.readEntryAt(-1)
-	if err != nil {
-		return nil, false, err
-	}
-
-	last := e
-	found := false
-	for pos > 0 {
-		key, _ := e.KeyValue()
-		switch bytes.Compare(key, needle) {
-		case 1: // key=ccc, needle=bbb
-			return last, found, nil
-		case 0: // equal
-			if first {
-				return e, true, nil
-			}
-			found = true
-			fallthrough
-		case -1: // key=aaa, needle=bbb
-			last = e
-			e, pos, err = s.readEntryAt(pos)
-			if err != nil {
-				return nil, found, err
-			}
-		}
-	}
-	return last, found, nil
-}
-func (s *segmentReader) readEntryAt(pos int64) (*entryBytes, int64, error) {
-	if pos < 0 {
-		pos = s.head.end
-	}
-	head := make([]byte, binary.MaxVarintLen16)
-	s.rd.ReadAt(head, pos-binary.MaxVarintLen16)
-	length, hsize := binary.Uvarint(reverse(head))
-
-	e := make(entryBytes, length)
-	_, err := s.rd.ReadAt(e, pos-int64(length)-int64(hsize))
-
-	return &e, pos - int64(length) - int64(hsize), err
 }
 
 type logFile struct {
-	rd interface {
+	header
+	Commit
+}
+
+func (h *logFile) AppendMagic(data []byte) []byte {
+	size := len(data)
+	data = append(data, Magic[:]...)
+	if h.end == 0 {
+		return data
+	}
+	hdata := binary.AppendUvarint(make([]byte, 0, 10), h.end)
+	reverse(hdata)
+	copy(data[size+10-len(hdata):], hdata)
+
+	return data
+}
+
+// WriteLogFile writes a log file to w, given a list of segments.
+// The caller is responsible for calling WriteAt on the correct offset.
+// The function will return an error if any of the segments fail to write.
+// The offset is the initial offset of the first segment, and will be
+// incremented by the length of the segment on each write.
+//
+// The log file is written with the following format:
+//   - A header with the magic, version, and flag (Dirty)
+//   - A series of segments, each with:
+//   - A footer with the length and hash of the segment
+//   - The contents of the segment
+//   - A header with the magic, version, flag (Clean), and end offset
+func WriteLogFile(w io.WriterAt, segments iter.Seq[io.Reader]) error {
+	_, err := w.WriteAt(Magic[:], 0)
+	if err != nil {
+		return err
+	}
+
+	lf := &LogWriter{
+		WriterAt: w,
+	}
+
+	return lf.writeIter(segments)
+}
+
+type rw interface {
 	io.ReaderAt
-		io.WriterTo
+	io.WriterAt
-	}
-	segments []segmentReader
-
-	fs.File
 }
 
-func ReadFile(fd fs.File) (*logFile, error) {
+func AppendLogFile(rw rw, segments iter.Seq[io.Reader]) error {
-	l := &logFile{File: fd}
+	logFile, err := ReadLogFile(rw)
-
-	stat, err := fd.Stat()
 	if err != nil {
-		return nil, err
+		return err
+	}
+	lf := &LogWriter{
+		WriterAt: rw,
+		logFile:  logFile.logFile,
+	}
+	return lf.writeIter(segments)
 }
 
-	eof := stat.Size()
+func (lf *LogWriter) writeIter(segments iter.Seq[io.Reader]) error {
-	if rd, ok := fd.(interface {
+	lf.size = 0
+	for s := range segments {
+		n, err := lf.writeBlock(s)
+		if err != nil {
+			return err
+		}
+		lf.end += n
+		lf.size += n
+		lf.count++
+	}
+
+	// Write the footer to the log file.
+	// The footer is written at the current end of file position.
+	n, err := lf.WriteAt(lf.AppendTrailer(make([]byte, 0, maxCommitSize)), int64(lf.end)+10)
+	if err != nil {
+		// If there is an error, return it.
+		return err
+	}
+	lf.end += uint64(n)
+
+	_, err = lf.WriteAt(lf.AppendMagic(make([]byte, 0, 10)), 0)
+
+	return err
+}
+
+type LogWriter struct {
+	logFile
+	io.WriterAt
+}
+
+// writeBlock writes a segment to the log file at the current end of file position.
+// The segment is written in chunks of 1024 bytes, and the hash of the segment
+func (lf *LogWriter) writeBlock(segment io.Reader) (uint64, error) {
+	h := hash()
+	block := Block{}
+
+	start := int64(lf.end) + 10
+	end := start
+
+	bytesWritten := 0
+
+	// Write the header to the log file.
+	// The footer is written at the current end of file position.
+	n, err := lf.WriteAt(make([]byte, headerSize), start)
+	bytesWritten += n
+	end += int64(n)
+	if err != nil {
+		// If there is an error, return it.
+		return uint64(bytesWritten), err
+	}
+
+	// Write the segment to the log file.
+	// The segment is written in chunks of 1024 bytes.
+	for {
+		// Read a chunk of the segment.
+		buf := make([]byte, 1024)
+		n, err := segment.Read(buf)
+		if err != nil {
+			// If the segment is empty, break the loop.
+			if err == io.EOF {
+				break
+			}
+			// If there is an error, return it.
+			return uint64(bytesWritten), err
+		}
+
+		// Compute the hash of the chunk.
+		h.Write(buf[:n])
+
+		// Write the chunk to the log file.
+		// The chunk is written at the current end of file position.
+		n, err = lf.WriteAt(buf[:n], end)
+		bytesWritten += n
+		if err != nil {
+			// If there is an error, return it.
+			return uint64(bytesWritten), err
+		}
+
+		// Update the length of the segment.
+		end += int64(n)
+		block.size += uint64(n)
+	}
+
+	block.extra = h.Sum(nil)
+	block.end += block.size
+
+	// Write the footer to the log file.
+	// The footer is written at the current end of file position.
+	n, err = lf.WriteAt(block.AppendTrailer(make([]byte, 0, maxBlockSize)), end)
+	bytesWritten += n
+	if err != nil {
+		// If there is an error, return it.
+		return uint64(bytesWritten), err
+	}
+	end += int64(n)
+	block.end += uint64(n)
+
+	// Update header to the log file.
+	// The footer is written at the current end of file position.
+	_, err = lf.WriteAt(block.AppendHeader(make([]byte, 0, headerSize)), start)
+	if err != nil {
+		// If there is an error, return it.
+		return uint64(bytesWritten), err
+	}
+
+	return uint64(bytesWritten), nil
+}
+
+// reverse reverses a slice in-place.
+func reverse[T any](b []T) {
+	l := len(b)
+	for i := 0; i < l/2; i++ {
+		b[i], b[l-i-1] = b[l-i-1], b[i]
+	}
+}
+
+type LogReader struct {
+	logFile
 	io.ReaderAt
-		io.WriterTo
+	Err error
-	}); ok {
+}
-		l.rd = rd
 
-	} else {
+// ReadLogFile reads a log file from the given io.ReaderAt. It returns a pointer to a LogFile, or an error if the file
-		rd, err := io.ReadAll(fd)
+// could not be read.
+func ReadLogFile(reader io.ReaderAt) (*LogReader, error) {
+	header := make([]byte, headerSize)
+	n, err := rsr(reader, 0, 10).ReadAt(header, 0)
 	if err != nil {
 		return nil, err
 	}
-		l.rd = bytes.NewReader(rd)
+	header = header[:n]
-	}
 
-	head := make([]byte, segmentFooterLength)
+	logFile := &LogReader{ReaderAt: reader}
-	for eof > 0 {
+	err = logFile.header.UnmarshalBinary(header)
-		_, err = l.rd.ReadAt(head, eof-int64(segmentFooterLength))
 	if err != nil {
 		return nil, err
 	}
 
-		s := segmentReader{
+	if logFile.end == 0 {
-			rd: l.rd,
+		return logFile, nil
 	}
-		s.head, err = ReadHead(head)
+
-		s.head.end = eof - int64(s.head.headlen)
+	commit := make([]byte, maxCommitSize)
-		if err != nil {
+	n, err = rsr(reader, 10, int64(logFile.end)).ReadAt(commit, 0)
+	if n == 0 && err != nil {
 		return nil, err
 	}
-		eof -= int64(s.head.datalen) + int64(s.head.headlen)
+	commit = commit[:n]
-		l.segments = append(l.segments, s)
+
+	err = logFile.Commit.UnmarshalBinary(commit)
+
+	return logFile, err
 }
 
-	return l, nil
+// Iterate reads the log file and calls the given function for each segment.
+// It passes an io.Reader that reads from the current segment. It will stop
+// calling the function if the function returns false.
+func (lf *LogReader) Iter(begin uint64) iter.Seq2[blockInfo, io.Reader] {
+	var commits []*Commit
+	for commit := range lf.iterCommits() {
+		commits = append(commits, &commit)
+	}
+	if lf.Err != nil {
+		return func(yield func(blockInfo, io.Reader) bool) {}
 	}
 
-func (l *logFile) Count() int64 {
+	reverse(commits)
-	return int64(len(l.segments))
-}
-func (l *logFile) LoadSegment(pos int64) (*segmentBytes, error) {
-	if pos < 0 {
-		pos = int64(len(l.segments) - 1)
-	}
-	if pos > int64(len(l.segments)-1) {
-		return nil, ErrDecode
-	}
-	s := l.segments[pos]
 
-	b := make([]byte, s.head.datalen+s.head.headlen)
+	return func(yield func(blockInfo, io.Reader) bool) {
-	_, err := l.rd.ReadAt(b, s.head.end-int64(len(b)))
+		start := int64(10)
+		var adj uint64
+		for _, commit := range commits {
+			size := int64(commit.size)
+			it := iterBlocks(io.NewSectionReader(lf, start, size), size)
+			for bi, block := range it {
+				bi.Commit = *commit
+				bi.Index += adj
+				bi.Start += uint64(start)
+				if begin <= bi.Index {
+					if !yield(bi, block) {
+						return
+					}
+				}
+			}
+
+			start += size + int64(commit.tsize)
+			adj = commit.count
+		}
+	}
+}
+
+type blockInfo struct{
+	Index uint64
+	Commit Commit
+	Start uint64
+	Size uint64
+	Hash []byte
+}
+
+func iterBlocks(r io.ReaderAt, end int64) iter.Seq2[blockInfo, io.Reader] {
+	var start int64
+	var i uint64
+	var bi blockInfo
+
+	return func(yield func(blockInfo, io.Reader) bool) {
+		buf := make([]byte, maxBlockSize)
+		for start < end {
+			block := &Block{}
+			buf = buf[:10]
+			n, err := rsr(r, int64(start), 10).ReadAt(buf, 0)
+			if n == 0 && err != nil {
+				return
+			}
+			start += int64(n)
+
+			if err := block.header.UnmarshalBinary(buf); err != nil {
+				return
+			}
+			buf = buf[:maxBlockSize]
+			n, err = rsr(r, int64(start), int64(block.end)).ReadAt(buf, 0)
+			if n == 0 && err != nil {
+				return
+			}
+			buf = buf[:n]
+			err = block.UnmarshalBinary(buf)
 			if err != nil {
-		return nil, err
+				return
 			}
 
-	return &segmentBytes{b, -1}, nil
+			bi.Index = i
-}
+			bi.Start = uint64(start)
-func (l *logFile) Find(needle []byte, first bool) (*entryBytes, bool, error) {
+			bi.Size = block.size
-	var cur, last segmentReader
+			bi.Hash = block.extra
 
-	for _, s := range l.segments {
+			if !yield(bi, io.NewSectionReader(r, int64(start), int64(block.size))) {
-		cur = s
+				return
-		e, err := cur.FirstEntry()
+			}
+
+			i++
+			start += int64(block.end)
+		}
+	}
+}
+
+func (lf *LogReader) iterCommits() iter.Seq[Commit] {
+	if lf.end == 0 {
+		return slices.Values([]Commit(nil))
+	}
+
+	offset := lf.end - lf.size - uint64(lf.tsize)
+	return func(yield func(Commit) bool) {
+		if !yield(lf.Commit) {
+			return
+		}
+
+		buf := make([]byte, maxCommitSize)
+
+		for offset > 10 {
+			commit := Commit{}
+			buf = buf[:10]
+			n, err := rsr(lf, 10, int64(offset)).ReadAt(buf, 0)
+			if n == 0 && err != nil {
+				lf.Err = err
+				return
+			}
+			buf = buf[:n]
+			err = commit.UnmarshalBinary(buf)
 			if err != nil {
-			return nil, false, err
+				lf.Err = err
+				return
 			}
-		k, _ := e.KeyValue()
+			if !yield(commit) {
-
+				return
-		if first && bytes.Equal(k, needle) {
-			break
 			}
-		if first && bytes.Compare(k, needle) > 0 {
+			offset -= commit.size + uint64(commit.tsize)
-			e, ok, err := cur.Find(needle, first)
-			if ok || err != nil{
-				return e, ok, err
 		}
-			break
 	}
-		if !first && bytes.Compare(k, needle) > 0 {
-			break
-		}
-		last = s
 }
 
-	e, ok, err := last.Find(needle, first)
+func (lf *LogReader) Rev(begin uint64) iter.Seq2[blockInfo, io.Reader] {
-	if ok || err != nil{
+	end := lf.end + 10
-		return e, ok, err
+	bi := blockInfo{}
+	bi.Index = lf.count-1
+
+	return func(yield func(blockInfo, io.Reader) bool) {
+		buf := make([]byte, maxBlockSize)
+
+		for commit := range lf.iterCommits() {
+			end -= uint64(commit.tsize)
+			start := end - commit.size
+
+			bi.Commit = commit
+
+			for start < end {
+				block := &Block{}
+				buf = buf[:maxBlockSize]
+				n, err := rsr(lf, int64(start), int64(commit.size)).ReadAt(buf, 0)
+				if n == 0 && err != nil {
+					lf.Err = err
+					return
 				}
-	// if by mistake it was not found in the last.. check the next segment.
+				buf = buf[:n]
-	return cur.Find(needle, first)
+				err = block.UnmarshalBinary(buf)
+				if err != nil {
+					lf.Err = err
+					return
 				}
-func (l *logFile) WriteTo(w io.Writer) (int64, error) {
+				if begin >= bi.Index {
-	return l.rd.WriteTo(w)
+
+					bi.Start = uint64(end-block.size)-uint64(block.tsize)
+					bi.Size = block.size
+
+					buf = buf[:10]
+					_, err = rsr(lf, int64(bi.Start)-10, 10).ReadAt(buf, 0)
+					if err != nil {
+						lf.Err = err
+						return
+					}
+					err = block.header.UnmarshalBinary(buf)
+					if err != nil {
+						lf.Err = err
+						return
 					}
 
-type segmentBytes struct {
+					bi.Hash = block.extra
-	b   []byte
+
-	pos int
+					if !yield(bi, io.NewSectionReader(lf, int64(bi.Start), int64(bi.Size))) {
+						return
+					}
+				}
+				end -= block.size + 10 + uint64(block.tsize)
+				bi.Index--
 			}
 
-type dataset struct {
+		}
-	rd    io.ReaderAt
+	}
-	files []logFile
-
-	fs.FS
 }
 
-func ReadDataset(fd fs.FS) (*dataset, error) {
+func (lf *LogReader) Count() uint64 {
-	panic("not implemented")
+	return lf.count
+}
+
+func (lf *LogReader) Size() uint64 {
+	return lf.end + 10
+}
+
+func rsr(r io.ReaderAt, offset, size int64) *revSegmentReader {
+	r = io.NewSectionReader(r, offset, size)
+	return &revSegmentReader{r, size}
+}
+
+type revSegmentReader struct {
+	io.ReaderAt
+	size int64
+}
+
+func (r *revSegmentReader) ReadAt(data []byte, offset int64) (int, error) {
+	if offset < 0 {
+		return 0, errors.New("negative offset")
+	}
+
+	if offset > int64(r.size) {
+		return 0, io.EOF
+	}
+
+	o := r.size - int64(len(data)) - offset
+	d := int64(len(data))
+	if o < 0 {
+		d = max(0, d+o)
+	}
+
+	i, err := r.ReaderAt.ReadAt(data[:d], max(0, o))
+	reverse(data[:i])
+	return i, err
 }

lsm/sst_test.go

@@ -1,327 +1,300 @@
-// SPDX-FileCopyrightText: 2023 Jon Lundy <jon@xuu.cc>
-// SPDX-License-Identifier: BSD-3-Clause
-
 package lsm
 
 import (
 	"bytes"
-	crand "crypto/rand"
 	"encoding/base64"
+	"errors"
 	"io"
-	"io/fs"
+	"iter"
-	"math/rand"
+	"slices"
-	"os"
-	"sort"
-	"sync"
 	"testing"
-	"time"
 
+	"github.com/docopt/docopt-go"
 	"github.com/matryer/is"
 )
 
-func TestLargeFile(t *testing.T) {
+// TestWriteLogFile tests AppendLogFile and WriteLogFile against a set of test cases.
-	is := is.New(t)
+//
-
+// Each test case contains a slice of slices of io.Readers, which are passed to
-	segCount := 4098
+// AppendLogFile and WriteLogFile in order. The test case also contains the
-
+// expected encoded output as a base64 string, as well as the expected output
-	f := randFile(t, 2_000_000, segCount)
+// when the file is read back using ReadLogFile.
-
+//
-	sf, err := ReadFile(f)
+// The test case also contains the expected output when the file is read back in
-	is.NoErr(err)
+// reverse order using ReadLogFile.Rev().
-
+//
-	is.True(len(sf.segments) <= segCount)
+// The test cases are as follows:
-	var needle []byte
+//
-	for i, s := range sf.segments {
+// - nil reader: Passes a nil slice of io.Readers to WriteLogFile.
-		e, err := s.FirstEntry()
+// - err reader: Passes a slice of io.Readers to WriteLogFile which returns an
-		is.NoErr(err)
+//   error when read.
-		k, v := e.KeyValue()
+// - single reader: Passes a single io.Reader to WriteLogFile.
-		needle = k
+// - multiple readers: Passes a slice of multiple io.Readers to WriteLogFile.
-		t.Logf("Segment-%d: %s = %d", i, k, v)
+// - multiple commit: Passes multiple slices of io.Readers to AppendLogFile.
+// - multiple commit 3x: Passes multiple slices of io.Readers to AppendLogFile
+//   three times.
+//
+// The test uses the is package from github.com/matryer/is to check that the
+// output matches the expected output.
+func TestWriteLogFile(t *testing.T) {
+	type test struct {
+		name string
+		in   [][]io.Reader
+		enc  string
+		out  [][]byte
+		rev  [][]byte
 	}
-	t.Log(f.Stat())
+	tests := []test{
-
+		{
-	tt, ok, err := sf.Find(needle, true)
+			name: "nil reader",
-	is.NoErr(err)
+			in:   nil,
-	is.True(ok)
+			enc:  "U291ci5pcwAAAwACAA",
-	key, val := tt.KeyValue()
+			out:  [][]byte{},
-	t.Log(string(key), val)
+			rev:  [][]byte{},
-
-	tt, ok, err = sf.Find([]byte("needle"), false)
-	is.NoErr(err)
-	is.True(!ok)
-	key, val = tt.KeyValue()
-	t.Log(string(key), val)
-
-	tt, ok, err = sf.Find([]byte{'\xff'}, false)
-	is.NoErr(err)
-	is.True(!ok)
-	key, val = tt.KeyValue()
-	t.Log(string(key), val)
-}
-
-func TestLargeFileDisk(t *testing.T) {
-	is := is.New(t)
-
-	segCount := 4098
-
-	t.Log("generate large file")
-	f := randFile(t, 2_000_000, segCount)
-
-	fd, err := os.CreateTemp("", "sst*")
-	is.NoErr(err)
-	defer func() { t.Log("cleanup:", fd.Name()); fd.Close(); os.Remove(fd.Name()) }()
-
-	t.Log("write file:", fd.Name())
-	_, err = io.Copy(fd, f)
-	is.NoErr(err)
-	fd.Seek(0, 0)
-
-	sf, err := ReadFile(fd)
-	is.NoErr(err)
-
-	is.True(len(sf.segments) <= segCount)
-	var needle []byte
-	for i, s := range sf.segments {
-		e, err := s.FirstEntry()
-		is.NoErr(err)
-		k, v := e.KeyValue()
-		needle = k
-
-		ok, err := s.VerifyHash()
-		is.NoErr(err)
-
-		t.Logf("Segment-%d: %s = %d %t", i, k, v, ok)
-		is.True(ok)
-	}
-	t.Log(f.Stat())
-
-	tt, ok, err := sf.Find(needle, false)
-	is.NoErr(err)
-	is.True(ok)
-	key, val := tt.KeyValue()
-	t.Log(string(key), val)
-
-	tt, ok, err = sf.Find([]byte("needle"), false)
-	is.NoErr(err)
-	is.True(!ok)
-	key, val = tt.KeyValue()
-	t.Log(string(key), val)
-
-	tt, ok, err = sf.Find([]byte{'\xff'}, false)
-	is.NoErr(err)
-	is.True(!ok)
-	key, val = tt.KeyValue()
-	t.Log(string(key), val)
-}
-
-func BenchmarkLargeFile(b *testing.B) {
-	segCount := 4098 / 4
-	f := randFile(b, 2_000_000, segCount)
-
-	sf, err := ReadFile(f)
-	if err != nil {
-		b.Error(err)
-	}
-	key := make([]byte, 5)
-	keys := make([][]byte, b.N)
-	for i := range keys {
-		_, err = crand.Read(key)
-		if err != nil {
-			b.Error(err)
-		}
-		keys[i] = []byte(base64.RawURLEncoding.EncodeToString(key))
-	}
-	b.Log("ready", b.N)
-	b.ResetTimer()
-	okays := 0
-	each := b.N / 10
-	for n := 0; n < b.N; n++ {
-		if each > 0 && n%each == 0 {
-			b.Log(n)
-		}
-		_, ok, err := sf.Find(keys[n], false)
-		if err != nil {
-			b.Error(err)
-		}
-		if ok {
-			okays++
-		}
-	}
-	b.Log("okays=", b.N, okays)
-}
-
-// TestFindRange is an initial range find for start and stop of a range of needles.
-// TODO: start the second query from where the first left off. Use an iterator?
-func TestFindRange(t *testing.T) {
-	is := is.New(t)
-
-	f := basicFile(t, 
-		entries{
-			{"AD", 5},
-			{"AC", 5},
-			{"AB", 4},
-			{"AB", 3},
 		},
-		entries{
+		{
-			{"AB", 2},
+			name: "err reader",
-			{"AA", 1},
+			in:   nil,
+			enc:  "U291ci5pcwAAAwACAA",
+			out:  [][]byte{},
+			rev:  [][]byte{},
 		},
-	)
+		{
-	sf, err := ReadFile(f)
+			name: "single reader",
-	is.NoErr(err)
+			in: [][]io.Reader{
-
+				{
-	var ok bool
+					bytes.NewBuffer([]byte{1, 2, 3, 4})}},
-	var first, last  *entryBytes
+			enc: "U291ci5pcwAAE756XndRZXhdAAYBAgMEAQQBAhA",
-
+			out: [][]byte{{1, 2, 3, 4}},
-	first, ok, err = sf.Find([]byte("AB"), true)
+			rev: [][]byte{{1, 2, 3, 4}}},
-	is.NoErr(err)
+		{
-
+			name: "multiple readers",
-	key, val := first.KeyValue()
+			in: [][]io.Reader{
-	t.Log(string(key), val)
+				{
-
+					bytes.NewBuffer([]byte{1, 2, 3, 4}),
-	is.True(ok)
+					bytes.NewBuffer([]byte{5, 6, 7, 8})}},
-	is.Equal(key, []byte("AB"))
+			enc: "U291ci5pcwAAI756XndRZXhdAAYBAgMEAQRhQyZWDDn5BQAGBQYHCAEEAgIg",
-	is.Equal(val, uint64(2))
+			out: [][]byte{{1, 2, 3, 4}, {5, 6, 7, 8}},
-
+			rev: [][]byte{{5, 6, 7, 8}, {1, 2, 3, 4}}},
-	last, ok, err = sf.Find([]byte("AB"), false)
+		{
-	is.NoErr(err)
+			name: "multiple commit",
-
+			in: [][]io.Reader{
-	key, val = last.KeyValue()
+				{
-	t.Log(string(key), val)
+					bytes.NewBuffer([]byte{1, 2, 3, 4})},
-
+				{
-	is.True(ok)
+					bytes.NewBuffer([]byte{5, 6, 7, 8})}},
-	is.Equal(key, []byte("AB"))
+			enc: "U291ci5pcwAAJr56XndRZXhdAAYBAgMEAQQBAhBhQyZWDDn5BQAGBQYHCAEEAgIQ",
-	is.Equal(val, uint64(4))
+			out: [][]byte{{1, 2, 3, 4}, {5, 6, 7, 8}},
-
+			rev: [][]byte{{5, 6, 7, 8}, {1, 2, 3, 4}}},
-
+		{
-	last, ok, err = sf.Find([]byte("AC"), false)
+			name: "multiple commit",
-	is.NoErr(err)
+			in: [][]io.Reader{
-
+				{
-	key, val = last.KeyValue()
+					bytes.NewBuffer([]byte{1, 2, 3, 4}),
-	t.Log(string(key), val)
+					bytes.NewBuffer([]byte{5, 6, 7, 8})},
-
+				{
-	is.True(ok)
+					bytes.NewBuffer([]byte{9, 10, 11, 12})},
-	is.Equal(key, []byte("AC"))
+			},
-	is.Equal(val, uint64(5))
+			enc: "U291ci5pcwAANr56XndRZXhdAAYBAgMEAQRhQyZWDDn5BQAGBQYHCAEEAgIgA4Buuio8Ro0ABgkKCwwBBAMCEA",
+			out: [][]byte{{1, 2, 3, 4}, {5, 6, 7, 8}, {9, 10, 11, 12}},
+			rev: [][]byte{{9, 10, 11, 12}, {5, 6, 7, 8}, {1, 2, 3, 4}}},
+		{
+			name: "multiple commit 3x",
+			in: [][]io.Reader{
+				{
+					bytes.NewBuffer([]byte{1, 2, 3}),
+					bytes.NewBuffer([]byte{4, 5, 6}),
+				},
+				{
+					bytes.NewBuffer([]byte{7, 8, 9}),
+				},
+				{
+					bytes.NewBuffer([]byte{10, 11, 12}),
+					bytes.NewBuffer([]byte{13, 14, 15}),
+				},
+			},
+			enc: "U291ci5pcwAAVNCqYhhnLPWrAAUBAgMBA7axWhhYd+HsAAUEBQYBAwICHr9ryhhdbkEZAAUHCAkBAwMCDy/UIhidCwCqAAUKCwwBA/NCwhh6wXgXAAUNDg8BAwUCHg",
+			out: [][]byte{{1, 2, 3}, {4, 5, 6}, {7, 8, 9}, {10, 11, 12}, {13, 14, 15}},
+			rev: [][]byte{{13, 14, 15}, {10, 11, 12}, {7, 8, 9}, {4, 5, 6}, {1, 2, 3}}},
 	}	
 
-func randFile(t interface {
+	for _, test := range tests {
-	Helper()
+		t.Run(test.name, func(t *testing.T) {
-	Error(...any)
+			is := is.New(t)
-}, size int, segments int) fs.File {
+			buf := &buffer{}
-	t.Helper()
 
-	lis := make(listEntries, size)
+			buffers := 0
-	for i := range lis {
+
-		key := make([]byte, 5)
+			if len(test.in) == 0 {
-		_, err := crand.Read(key)
+				err := WriteLogFile(buf, slices.Values([]io.Reader{}))
-		if err != nil {
+				is.NoErr(err)
-			t.Error(err)
 			}
-		key = []byte(base64.RawURLEncoding.EncodeToString(key))
+			for i, in := range test.in {
-		// key := []byte(fmt.Sprintf("key-%05d", i))
+				buffers += len(in)
 
-		lis[i] = NewKeyValue(key, rand.Uint64()%16_777_216)
+				if i == 0 {
+					err := WriteLogFile(buf, slices.Values(in))
+					is.NoErr(err)
+				} else {
+					err := AppendLogFile(buf, slices.Values(in))
+					is.NoErr(err)
+				}
 			}
 			
-	sort.Sort(sort.Reverse(&lis))
+			is.Equal(base64.RawStdEncoding.EncodeToString(buf.Bytes()), test.enc)
-	each := size / segments
-	if size%segments != 0 {
-		each++
-	}
-	split := make([]listEntries, segments)
 
-	for i := range split {
+			files, err := ReadLogFile(bytes.NewReader(buf.Bytes()))
-		if (i+1)*each > len(lis) {
+			is.NoErr(err)
-			split[i] = lis[i*each : i*each+len(lis[i*each:])]
+
-			split = split[:i+1]
+			is.Equal(files.Size(), uint64(len(buf.Bytes())))
+
+			i := 0
+			for bi, fp := range files.Iter(0) {
+				buf, err := io.ReadAll(fp)
+				is.NoErr(err)
+
+				hash := hash()
+				hash.Write(buf)
+				is.Equal(bi.Hash, hash.Sum(nil)[:len(bi.Hash)])
+
+				is.True(len(test.out) > int(bi.Index))
+				is.Equal(buf, test.out[bi.Index])
+				i++
+			}
+			is.NoErr(files.Err)
+			is.Equal(i, buffers)
+
+			i = 0
+			for bi, fp := range files.Rev(files.Count()) {
+				buf, err := io.ReadAll(fp)
+				is.NoErr(err)
+
+				hash := hash()
+				hash.Write(buf)
+				is.Equal(bi.Hash, hash.Sum(nil)[:len(bi.Hash)])
+
+				is.Equal(buf, test.rev[i])
+				is.Equal(buf, test.out[bi.Index])
+				i++
+			}
+			is.NoErr(files.Err)
+			is.Equal(i, buffers)
+			is.Equal(files.Count(), uint64(i))
+		})
+	}
+}
+
+// TestArgs tests that the CLI arguments are correctly parsed.
+func TestArgs(t *testing.T) {
+	is := is.New(t)
+	usage := `Usage: lsm2 create <archive> <files>...`
+
+	arguments, err := docopt.ParseArgs(usage, []string{"create", "archive", "file1", "file2"}, "1.0")
+	is.NoErr(err)
+
+	var params struct {
+		Create  bool     `docopt:"create"`
+		Archive string   `docopt:"<archive>"`
+		Files   []string `docopt:"<files>"`
+	}
+	err = arguments.Bind(&params)
+	is.NoErr(err)
+
+	is.Equal(params.Create, true)
+	is.Equal(params.Archive, "archive")
+	is.Equal(params.Files, []string{"file1", "file2"})
+}
+
+func BenchmarkIterate(b *testing.B) {
+	block := make([]byte, 1024)
+	buf := &buffer{}
+
+
+	b.Run("write", func(b *testing.B) {
+		WriteLogFile(buf, func(yield func(io.Reader) bool) {
+			for range (b.N) {
+				if !yield(bytes.NewBuffer(block)) {
 					break	
 				}	
-		split[i] = lis[i*each : (i+1)*each]
+			}
+		})
+	})
+
+	b.Run("read", func(b *testing.B) {
+		lf, _ := ReadLogFile(buf)
+		b.Log(lf.Count())
+		for range (b.N) {
+			for _, fp := range lf.Iter(0) {
+				_, _ = io.Copy(io.Discard, fp)
+				break
+			}
+		}
+	})
+
+	b.Run("rev", func(b *testing.B) {
+		lf, _ := ReadLogFile(buf)
+		b.Log(lf.Count())
+		for range (b.N) {
+			for _, fp := range lf.Rev(lf.Count()) {
+				_, _ = io.Copy(io.Discard, fp)
+				break
+			}
+		}
+	})
 }
 
-	var b bytes.Buffer
+type buffer struct {
-	for _, s := range split {
+	buf []byte
-		s.WriteTo(&b)
 }
 
-	return NewFile(b.Bytes())
+// Bytes returns the underlying byte slice of the bufferWriterAt.
+func (b *buffer) Bytes() []byte {
+	return b.buf
 }
 
-type fakeStat struct {
+// WriteAt implements io.WriterAt. It appends data to the internal buffer
-	size int64
+// if the offset is beyond the current length of the buffer. It will
+// return an error if the offset is negative.
+func (b *buffer) WriteAt(data []byte, offset int64) (written int, err error) {
+	if offset < 0 {
+		return 0, errors.New("negative offset")
 	}
 
-// IsDir implements fs.FileInfo.
+	currentLength := int64(len(b.buf))
-func (*fakeStat) IsDir() bool {
+	if currentLength < offset+int64(len(data)) {
-	panic("unimplemented")
+		b.buf = append(b.buf, make([]byte, offset+int64(len(data))-currentLength)...)
 	}
 
-// ModTime implements fs.FileInfo.
+	written = copy(b.buf[offset:], data)
-func (*fakeStat) ModTime() time.Time {
+	return
-	panic("unimplemented")
 }
 
-// Mode implements fs.FileInfo.
+// ReadAt implements io.ReaderAt. It reads data from the internal buffer starting
-func (*fakeStat) Mode() fs.FileMode {
+// from the specified offset and writes it into the provided data slice. If the
-	panic("unimplemented")
+// offset is negative, it returns an error. If the requested read extends beyond
+// the buffer's length, it returns the data read so far along with an io.EOF error.
+func (b *buffer) ReadAt(data []byte, offset int64) (int, error) {
+	if offset < 0 {
+		return 0, errors.New("negative offset")
 	}
 
-// Name implements fs.FileInfo.
+	if offset > int64(len(b.buf)) || len(b.buf[offset:]) < len(data) {
-func (*fakeStat) Name() string {
+		return copy(data, b.buf[offset:]), io.EOF
-	panic("unimplemented")
 	}
 
-// Size implements fs.FileInfo.
+	return copy(data, b.buf[offset:]), nil
-func (s *fakeStat) Size() int64 {
-	return s.size
 }
 
-// Sys implements fs.FileInfo.
+// IterOne takes an iterator that yields values of type T along with a value of
-func (*fakeStat) Sys() any {
+// type I, and returns an iterator that yields only the values of type T. It
-	panic("unimplemented")
+// discards the values of type I.
+func IterOne[I, T any](it iter.Seq2[I, T]) iter.Seq[T] {
+	return func(yield func(T) bool) {
+		for i, v := range it {
+			_ = i
+			if !yield(v) {
+				return
 			}
-
-var _ fs.FileInfo = (*fakeStat)(nil)
-
-type rd interface {
-	io.ReaderAt
-	io.Reader
 		}
-type fakeFile struct {
-	stat func() fs.FileInfo
-
-	rd
 	}
-
-func (fakeFile) Close() error                 { return nil }
-func (f fakeFile) Stat() (fs.FileInfo, error) { return f.stat(), nil }
-
-func NewFile(b ...[]byte) fs.File {
-	in := bytes.Join(b, nil)
-	rd := bytes.NewReader(in)
-	size := int64(len(in))
-	return &fakeFile{stat: func() fs.FileInfo { return &fakeStat{size: size} }, rd: rd}
 }
-func NewFileFromReader(rd *bytes.Reader) fs.File {
-	return &fakeFile{stat: func() fs.FileInfo { return &fakeStat{size: int64(rd.Len())} }, rd: rd}
-}
-
-type fakeFS struct {
-	files map[string]*fakeFile
-	mu    sync.RWMutex
-}
-
-// Open implements fs.FS.
-func (f *fakeFS) Open(name string) (fs.File, error) {
-	f.mu.RLock()
-	defer f.mu.RUnlock()
-
-	if file, ok := f.files[name]; ok {
-		return file, nil
-	}
-
-	return nil, fs.ErrNotExist
-}
-
-var _ fs.FS = (*fakeFS)(nil)

lsm2/cli/main.go

@@ -1,120 +0,0 @@
-package main
-
-import (
-	"errors"
-	"fmt"
-	"io"
-	"iter"
-	"os"
-
-	"github.com/docopt/docopt-go"
-	"go.sour.is/pkg/lsm2"
-)
-
-var usage = `
-Usage: 
-  lsm2 create <archive> <files>...
-  lsm2 append <archive> <files>...
-  lsm2 read <archive> <index>`
-
-type args struct {
-	Create  bool
-	Append  bool
-	Read    bool
-
-	Archive string   `docopt:"<archive>"`
-	Files   []string `docopt:"<files>"`
-	Index   int64  `docopt:"<index>"`
-}
-
-func main() {
-	opts, err := docopt.ParseDoc(usage)
-	if err != nil {
-		fmt.Println(err)
-		os.Exit(1)
-	}
-	fmt.Println(opts, must(opts.Bool("create")))
-	console := console{os.Stdin, os.Stdout, os.Stderr}
-	args := args{}
-	err = opts.Bind(&args)
-	fmt.Println(err)
-	run(console, args)
-}
-
-type console struct {
-	Stdin  io.Reader
-	Stdout io.Writer
-	Stderr io.Writer
-}
-
-func (c console) Write(b []byte) (int, error) {
-	return c.Stdout.Write(b)
-}
-
-func run(console console,a args) error {
-	fmt.Fprintln(console, "lsm")
-	switch {
-	case a.Create:
-		f, err := os.OpenFile(a.Archive, os.O_CREATE|os.O_WRONLY, 0644)
-		if err != nil {
-			return err
-		}
-		defer f.Close()
-
-		return lsm2.WriteLogFile(f, fileReaders(a.Files))
-	case a.Append:
-		f, err := os.OpenFile(a.Archive, os.O_RDWR, 0644)
-		if err != nil {
-			return err
-		}
-		defer f.Close()
-
-		return lsm2.AppendLogFile(f, fileReaders(a.Files))
-	case a.Read:
-		fmt.Fprintln(console, "reading", a.Archive)
-
-		f, err := os.Open(a.Archive)
-		if err != nil {
-			return err
-		}
-		defer f.Close()	
-		lg, err := lsm2.ReadLogFile(f)
-		if err != nil {
-			return err
-		}
-		for i, rd := range lg.Iter() {
-			fmt.Fprintf(console, "=========================\n%d:\n", i)
-			io.Copy(console, rd)
-			fmt.Fprintln(console, "=========================")
-		}
-		if lg.Err != nil {
-			return lg.Err
-		}
-		return nil
-	default:
-		return errors.New("unknown command")
-	}
-}
-
-func fileReaders(names []string) iter.Seq[io.Reader] {
-	return iter.Seq[io.Reader](func(yield func(io.Reader) bool) {
-		for _, name := range names {
-			f, err := os.Open(name)
-			if err != nil {
-				continue
-			}
-			if !yield(f) {
-				f.Close()
-				return
-			}
-			f.Close()
-		}
-	})
-}
-
-func must[T any](v T, err error) T {
-	if err != nil {
-		panic(err)
-	}
-	return v
-}

lsm2/sst.go

@@ -1,581 +0,0 @@
-package lsm2
-
-import (
-	"encoding/binary"
-	"errors"
-	"fmt"
-	"hash/fnv"
-	"io"
-	"iter"
-)
-
-// [Sour.is|size] [size|hash][data][hash|flag|size]... [prev|count|flag|size]
-
-// Commit1: [magic>|<end]{10} ... [<count][<size][<flag]{3..30}
-//                 +---------|--------------------------------> end = seek to end of file
-//                       <---|-------------+                    size = seek to magic header
-//                       <---|-------------+10                  size + 10 = seek to start of file
-//          <-----------------------------T+10----------------> 10 + size + trailer = full file size
-
-// Commit2: [magic>|<end]{10} ... [<count][<size][<flag]{3..30} ... [<prev][<count][<size][<flag]{4..40}
-//                           <---|---------+
-//                           <-------------+T----------------->
-//                  +--------|------------------------------------------------------------------------->
-//                           <-------------------------------------|----------------+
-//     prev = seek to last commit                              <---|-+
-//     prev + trailer = size of commit                         <----T+--------------------------------->
-
-// Block:  [hash>|<end]{10} ... [<size][<flag]{2..20}
-//               +---------|------------------------>  end = seek to end of block
-//                         <---|-+                     size = seek to end of header
-//         <-------------------|-+10                   size + 10 = seek to start of block
-//         <---------------------T+10--------------->  size + 10 + trailer = full block size
-
-const (
-	TypeUnknown uint64 = iota
-	TypeSegment
-	TypeCommit
-	TypePrevCommit
-
-	headerSize = 10
-
-	maxCommitSize = 4 * binary.MaxVarintLen64
-	minCommitSize = 3
-
-	maxBlockSize = 2 * binary.MaxVarintLen64
-	minBlockSize = 2
-)
-
-var (
-	Magic   = [10]byte([]byte("Sour.is\x00\x00\x00"))
-	Version = uint8(1)
-	hash    = fnv.New64a
-
-	ErrDecode = errors.New("decode")
-)
-
-type header struct {
-	end   uint64
-	extra []byte
-}
-
-// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-// It decodes the input binary data into the header struct.
-// The function expects the input data to be of a specific size (headerSize),
-// otherwise it returns an error indicating bad data.
-// It reads the 'end' field from the binary data, updates the 'extra' field,
-// and reverses the byte order of 'extra' in place.
-func (h *header) UnmarshalBinary(data []byte) error {
-	if len(data) != headerSize {
-		return fmt.Errorf("%w: bad data", ErrDecode)
-	}
-
-	h.extra = make([]byte, headerSize)
-	copy(h.extra, data)
-
-	var bytesRead int
-	h.end, bytesRead = binary.Uvarint(h.extra)
-	reverse(h.extra)
-	h.extra = h.extra[:headerSize-bytesRead]
-
-	return nil
-}
-
-type Commit struct {
-	flag  uint64 // flag values
-	size  uint64 // size of the trailer
-	count uint64 // number of entries
-	prev  uint64 // previous commit
-
-	tsize int
-}
-
-// Append marshals the trailer into binary form and appends it to data.
-// It returns the new slice.
-func (h *Commit) AppendTrailer(data []byte) []byte {
-	h.flag |= TypeCommit
-	// if h.prev > 0 {
-	// 	h.flag |= TypePrevCommit
-	// }
-
-	size := len(data)
-	data = binary.AppendUvarint(data, h.size)
-	data = binary.AppendUvarint(data, h.flag)
-	data = binary.AppendUvarint(data, h.count)
-	// if h.prev > 0 {
-	// 	data = binary.AppendUvarint(data, h.prev)
-	// }
-	reverse(data[size:])
-
-	return data
-}
-
-// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-// It reads a trailer from binary data, and sets the fields
-// of the receiver to the values found in the header.
-func (h *Commit) UnmarshalBinary(data []byte) error {
-	if len(data) < minCommitSize {
-		return fmt.Errorf("%w: bad data", ErrDecode)
-	}
-
-	var n int
-	h.size, n = binary.Uvarint(data)
-	data = data[n:]
-	h.tsize += n
-
-	h.flag, n = binary.Uvarint(data)
-	data = data[n:]
-	h.tsize += n
-
-	h.count, n = binary.Uvarint(data)
-	data = data[n:]
-	h.tsize += n
-
-	// h.prev = h.size
-	if h.flag&TypePrevCommit == TypePrevCommit {
-		h.prev, n = binary.Uvarint(data)
-		h.tsize += n
-	}
-
-	return nil
-}
-
-type Block struct {
-	header
-
-	size uint64
-	flag uint64
-
-	tsize int
-}
-
-func (h *Block) AppendHeader(data []byte) []byte {
-	size := len(data)
-	data = append(data, make([]byte, 10)...)
-	copy(data, h.extra)
-	if h.size == 0 {
-		return data
-	}
-	hdata := binary.AppendUvarint(make([]byte, 0, 10), h.end)
-	reverse(hdata)
-	copy(data[size+10-len(hdata):], hdata)
-
-	return data
-}
-
-// AppendTrailer marshals the footer into binary form and appends it to data.
-// It returns the new slice.
-func (h *Block) AppendTrailer(data []byte) []byte {
-	size := len(data)
-
-	h.flag |= TypeSegment
-	data = binary.AppendUvarint(data, h.size)
-	data = binary.AppendUvarint(data, h.flag)
-	reverse(data[size:])
-
-	return data
-}
-
-// UnmarshalBinary implements encoding.BinaryUnmarshaler.
-// It reads a footer from binary data, and sets the fields
-// of the receiver to the values found in the footer.
-func (h *Block) UnmarshalBinary(data []byte) error {
-	if len(data) < minBlockSize {
-		return fmt.Errorf("%w: bad data", ErrDecode)
-	}
-
-	var n int
-	h.size, n = binary.Uvarint(data)
-	data = data[n:]
-	h.tsize += n
-
-	h.flag, n = binary.Uvarint(data)
-	data = data[n:]
-	h.tsize += n
-
-	copy(h.extra, data[:8])
-
-	return nil
-}
-
-type logFile struct {
-	header
-	Commit
-}
-
-func (h *logFile) AppendMagic(data []byte) []byte {
-	size := len(data)
-	data = append(data, Magic[:]...)
-	if h.end == 0 {
-		return data
-	}
-	hdata := binary.AppendUvarint(make([]byte, 0, 10), h.end)
-	reverse(hdata)
-	copy(data[size+10-len(hdata):], hdata)
-
-	return data
-}
-
-// WriteLogFile writes a log file to w, given a list of segments.
-// The caller is responsible for calling WriteAt on the correct offset.
-// The function will return an error if any of the segments fail to write.
-// The offset is the initial offset of the first segment, and will be
-// incremented by the length of the segment on each write.
-//
-// The log file is written with the following format:
-//   - A header with the magic, version, and flag (Dirty)
-//   - A series of segments, each with:
-//   - A footer with the length and hash of the segment
-//   - The contents of the segment
-//   - A header with the magic, version, flag (Clean), and end offset
-func WriteLogFile(w io.WriterAt, segments iter.Seq[io.Reader]) error {
-	_, err := w.WriteAt(Magic[:], 0)
-	if err != nil {
-		return err
-	}
-
-	lf := &LogWriter{
-		WriterAt: w,
-	}
-
-	return lf.writeIter(segments)
-}
-
-type rw interface {
-	io.ReaderAt
-	io.WriterAt
-}
-
-func AppendLogFile(rw rw, segments iter.Seq[io.Reader]) error {
-	logFile, err := ReadLogFile(rw)
-	if err != nil {
-		return err
-	}
-	lf := &LogWriter{
-		WriterAt: rw,
-		logFile:  logFile.logFile,
-	}
-	return lf.writeIter( segments)
-}
-
-
-func (lf *LogWriter) writeIter(segments iter.Seq[io.Reader]) error {	
-	lf.size = 0
-	for s := range segments {
-		n, err := lf.writeBlock(s)
-		if err != nil {
-			return err
-		}
-		lf.end += n
-		lf.size += n
-		lf.count++
-	}
-
-	// Write the footer to the log file.
-	// The footer is written at the current end of file position.
-	n, err := lf.WriteAt(lf.AppendTrailer(make([]byte, 0, maxCommitSize)), int64(lf.end)+10)
-	if err != nil {
-		// If there is an error, return it.
-		return err
-	}
-	lf.end += uint64(n)
-
-	_, err = lf.WriteAt(lf.AppendMagic(make([]byte, 0, 10)), 0)
-
-	return err
-}
-
-type LogWriter struct {
-	logFile
-	io.WriterAt
-}
-
-// writeBlock writes a segment to the log file at the current end of file position.
-// The segment is written in chunks of 1024 bytes, and the hash of the segment
-func (lf *LogWriter) writeBlock(segment io.Reader) (uint64, error) {
-	h := hash()
-	block := Block{}
-
-	start := int64(lf.end) + 10
-	end := start
-	
-	bytesWritten := 0
-
-	// Write the header to the log file.
-	// The footer is written at the current end of file position.
-	n, err := lf.WriteAt(make([]byte, headerSize), start)
-	bytesWritten += n
-	end += int64(n)
-	if err != nil {
-		// If there is an error, return it.
-		return uint64(bytesWritten), err
-	}
-
-	// Write the segment to the log file.
-	// The segment is written in chunks of 1024 bytes.
-	for {
-		// Read a chunk of the segment.
-		buf := make([]byte, 1024)
-		n, err := segment.Read(buf)
-		if err != nil {
-			// If the segment is empty, break the loop.
-			if err == io.EOF {
-				break
-			}
-			// If there is an error, return it.
-			return uint64(bytesWritten), err
-		}
-
-		// Compute the hash of the chunk.
-		h.Write(buf[:n])
-
-		// Write the chunk to the log file.
-		// The chunk is written at the current end of file position.
-		n, err = lf.WriteAt(buf[:n], end)
-		bytesWritten += n
-		if err != nil {
-			// If there is an error, return it.
-			return uint64(bytesWritten), err
-		}
-
-		// Update the length of the segment.
-		end += int64(n)
-		block.size += uint64(n)
-	}
-
-	block.extra = h.Sum(nil)
-	block.end += block.size
-
-	// Write the footer to the log file.
-	// The footer is written at the current end of file position.
-	n, err = lf.WriteAt(block.AppendTrailer(make([]byte, 0, maxBlockSize)), end)
-	bytesWritten += n
-	if err != nil {
-		// If there is an error, return it.
-		return uint64(bytesWritten), err
-	}
-	end += int64(n)
-	block.end += uint64(n)
-
-	// Update header to the log file.
-	// The footer is written at the current end of file position.
-	_, err = lf.WriteAt(block.AppendHeader(make([]byte, 0, headerSize)), start)
-	if err != nil {
-		// If there is an error, return it.
-		return uint64(bytesWritten), err
-	}
-
-	return uint64(bytesWritten), nil
-}
-
-// reverse reverses a slice in-place.
-func reverse[T any](b []T) {
-	l := len(b)
-	for i := 0; i < l/2; i++ {
-		b[i], b[l-i-1] = b[l-i-1], b[i]
-	}
-}
-
-type LogReader struct {
-	logFile
-	io.ReaderAt
-	Err error
-}
-
-// ReadLogFile reads a log file from the given io.ReaderAt. It returns a pointer to a LogFile, or an error if the file
-// could not be read.
-func ReadLogFile(reader io.ReaderAt) (*LogReader, error) {
-	header := make([]byte, headerSize)
-	n, err := rsr(reader, 0, 10).ReadAt(header, 0)
-	if err != nil {
-		return nil, err
-	}
-	header = header[:n]
-
-	logFile := &LogReader{ReaderAt: reader}
-	err = logFile.header.UnmarshalBinary(header)
-	if err != nil {
-		return nil, err
-	}
-
-	if logFile.end == 0 {
-		return logFile, nil
-	}
-
-	commit := make([]byte, maxCommitSize)
-	n, err = rsr(reader, 10, int64(logFile.end)).ReadAt(commit, 0)
-	if n == 0 && err != nil {
-		return nil, err
-	}
-	commit = commit[:n]
-
-	err = logFile.Commit.UnmarshalBinary(commit)
-
-	return logFile, err
-}
-
-// Iterate reads the log file and calls the given function for each segment.
-// It passes an io.Reader that reads from the current segment. It will stop
-// calling the function if the function returns false.
-func (lf *LogReader) Iter() iter.Seq2[uint64, io.Reader] {
-	var commits []*Commit
-	for commit := range lf.iterCommits() {
-		commits = append(commits, &commit)
-	}
-	if lf.Err != nil {
-		return func(yield func(uint64, io.Reader) bool) {}
-	}
-
-	reverse(commits)
-
-	return func(yield func(uint64, io.Reader) bool) {
-		start := int64(10)
-		var adj uint64
-		for _, commit := range commits {
-			size := int64(commit.size)
-			it := iterBlocks(io.NewSectionReader(lf, start, size), size)
-			for i, block := range it {
-				if !yield(adj+i, block) {
-					return
-				}
-			}
-
-			start += size + int64(commit.tsize)
-			adj = commit.count
-		}
-	}
-}
-
-func iterBlocks(r io.ReaderAt, end int64) iter.Seq2[uint64, io.Reader] {
-	var start int64
-	var i uint64
-	return func(yield func(uint64, io.Reader) bool) {
-		buf := make([]byte, maxBlockSize)
-		for start < end {
-			block := &Block{}
-			buf = buf[:10]
-			n, err := rsr(r, int64(start), 10).ReadAt(buf, 0)
-			if n == 0 && err != nil {
-				return
-			}
-			start += int64(n)
-
-			if err := block.header.UnmarshalBinary(buf); err != nil {
-				return
-			}
-			buf = buf[:maxBlockSize]
-			n, err = rsr(r, int64(start), int64(block.end)).ReadAt(buf, 0)
-			if n == 0 && err != nil {
-				return
-			}
-			buf = buf[:n]
-			err = block.UnmarshalBinary(buf)
-			if err != nil {
-				return
-			}
-
-			if !yield(i, io.NewSectionReader(r, int64(start), int64(block.size))) {
-				return
-			}
-
-			i++
-			start += int64(block.end)
-		}
-	}
-}
-
-func (lf *LogReader) iterCommits() iter.Seq[Commit] {
-	if lf.end == 0 {
-		return func(yield func(Commit) bool) {}
-	}
-
-	offset := lf.end - lf.size - uint64(lf.tsize)
-	return func(yield func(Commit) bool) {
-		if !yield(lf.Commit) {
-			return
-		}
-
-		for offset > 10 {
-			commit := Commit{}
-			buf := make([]byte, maxCommitSize)
-			n, err := rsr(lf, 10, int64(offset)).ReadAt(buf, 0)
-			if n == 0 && err != nil {
-				lf.Err = err
-				return
-			}
-			buf = buf[:n]
-			err = commit.UnmarshalBinary(buf)
-			if err != nil {
-				lf.Err = err
-				return
-			}
-			if !yield(commit) {
-				return
-			}
-			offset -= commit.size + uint64(commit.tsize)
-		}
-	}
-}
-
-func (lf *LogReader) Rev() iter.Seq2[uint64, io.Reader] {
-	end := lf.end + 10
-	i := lf.count
-	return func(yield func(uint64, io.Reader) bool) {
-
-		for commit := range lf.iterCommits() {
-			end -= uint64(commit.tsize)
-			start := end - commit.size
-			for start < end {
-				block := &Block{}
-				buf := make([]byte, maxBlockSize)
-				n, err := rsr(lf, int64(start), int64(commit.size)).ReadAt(buf, 0)
-				if n == 0 && err != nil {
-					lf.Err = err
-					return
-				}
-				buf = buf[:n]
-				err = block.UnmarshalBinary(buf)
-				if err != nil {
-					lf.Err = err
-					return
-				}
-				if !yield(i-1, io.NewSectionReader(lf, int64(end-block.size)-int64(block.tsize), int64(block.size))) {
-					return
-				}
-				end -= block.size + 10 + uint64(block.tsize)
-				i--
-			}
-
-		}
-	}
-}
-
-func rsr(r io.ReaderAt, offset, size int64) *revSegmentReader {
-	r = io.NewSectionReader(r, offset, size)
-	return &revSegmentReader{r, size}
-}
-
-type revSegmentReader struct {
-	io.ReaderAt
-	size int64
-}
-
-func (r *revSegmentReader) ReadAt(data []byte, offset int64) (int, error) {
-	if offset < 0 {
-		return 0, errors.New("negative offset")
-	}
-
-	if offset > int64(r.size) {
-		return 0, io.EOF
-	}
-
-	o := r.size - int64(len(data)) - offset
-	d := int64(len(data))
-	if o < 0 {
-		d = max(0, d+o)
-	}
-
-	i, err := r.ReaderAt.ReadAt(data[:d], max(0, o))
-	reverse(data[:i])
-	return i, err
-}

lsm2/sst_test.go

@@ -1,252 +0,0 @@
-package lsm2
-
-import (
-	"bytes"
-	"encoding/base64"
-	"errors"
-	"io"
-	"iter"
-	"slices"
-	"testing"
-
-	"github.com/docopt/docopt-go"
-	"github.com/matryer/is"
-)
-
-// TestAppend tests AppendLogFile and WriteLogFile against a set of test cases.
-//
-// Each test case contains a slice of slices of io.Readers, which are passed to
-// AppendLogFile and WriteLogFile in order. The test case also contains the
-// expected encoded output as a base64 string, as well as the expected output
-// when the file is read back using ReadLogFile.
-//
-// The test case also contains the expected output when the file is read back in
-// reverse order using ReadLogFile.Rev().
-//
-// The test cases are as follows:
-//
-// - nil reader: Passes a nil slice of io.Readers to WriteLogFile.
-// - err reader: Passes a slice of io.Readers to WriteLogFile which returns an
-//   error when read.
-// - single reader: Passes a single io.Reader to WriteLogFile.
-// - multiple readers: Passes a slice of multiple io.Readers to WriteLogFile.
-// - multiple commit: Passes multiple slices of io.Readers to AppendLogFile.
-// - multiple commit 3x: Passes multiple slices of io.Readers to AppendLogFile
-//   three times.
-//
-// The test uses the is package from github.com/matryer/is to check that the
-// output matches the expected output.
-func TestAppend(t *testing.T) {
-	type test struct {
-		name string
-		in   [][]io.Reader
-		enc  string
-		out  [][]byte
-		rev  [][]byte
-	}
-	tests := []test{
-		{
-			name: "nil reader",
-			in:   nil,
-			enc:  "U291ci5pcwAAAwACAA",
-			out:  [][]byte{},
-			rev:  [][]byte{},
-		},
-		{
-			name: "err reader",
-			in:   nil,
-			enc:  "U291ci5pcwAAAwACAA",
-			out:  [][]byte{},
-			rev:  [][]byte{},
-		},
-		{
-			name: "single reader",
-			in: [][]io.Reader{
-				{
-					bytes.NewBuffer([]byte{1, 2, 3, 4})}},
-			enc: "U291ci5pcwAAE756XndRZXhdAAYBAgMEAQQBAhA",
-			out: [][]byte{{1, 2, 3, 4}},
-			rev: [][]byte{{1, 2, 3, 4}}},
-		{
-			name: "multiple readers",
-			in: [][]io.Reader{
-				{
-					bytes.NewBuffer([]byte{1, 2, 3, 4}),
-					bytes.NewBuffer([]byte{5, 6, 7, 8})}},
-			enc: "U291ci5pcwAAI756XndRZXhdAAYBAgMEAQRhQyZWDDn5BQAGBQYHCAEEAgIg",
-			out: [][]byte{{1, 2, 3, 4}, {5, 6, 7, 8}},
-			rev: [][]byte{{5, 6, 7, 8}, {1, 2, 3, 4}}},
-		{
-			name: "multiple commit",
-			in: [][]io.Reader{
-				{
-					bytes.NewBuffer([]byte{1, 2, 3, 4})},
-				{
-					bytes.NewBuffer([]byte{5, 6, 7, 8})}},
-			enc: "U291ci5pcwAAJr56XndRZXhdAAYBAgMEAQQBAhBhQyZWDDn5BQAGBQYHCAEEAgIQ",
-			out: [][]byte{{1, 2, 3, 4}, {5, 6, 7, 8}},
-			rev: [][]byte{{5, 6, 7, 8}, {1, 2, 3, 4}}},
-		{
-			name: "multiple commit",
-			in: [][]io.Reader{
-				{
-					bytes.NewBuffer([]byte{1, 2, 3, 4}),
-					bytes.NewBuffer([]byte{5, 6, 7, 8})},
-				{
-					bytes.NewBuffer([]byte{9, 10, 11, 12})},
-			},
-			enc: "U291ci5pcwAANr56XndRZXhdAAYBAgMEAQRhQyZWDDn5BQAGBQYHCAEEAgIgA4Buuio8Ro0ABgkKCwwBBAMCEA",
-			out: [][]byte{{1, 2, 3, 4}, {5, 6, 7, 8}, {9, 10, 11, 12}},
-			rev: [][]byte{{9, 10, 11, 12}, {5, 6, 7, 8}, {1, 2, 3, 4}}},
-		{
-			name: "multiple commit 3x",
-			in: [][]io.Reader{
-				{
-					bytes.NewBuffer([]byte{1, 2, 3}),
-					bytes.NewBuffer([]byte{4, 5, 6}),
-				},
-				{
-					bytes.NewBuffer([]byte{7, 8, 9}),
-				},
-				{
-					bytes.NewBuffer([]byte{10, 11, 12}),
-					bytes.NewBuffer([]byte{13, 14, 15}),
-				},
-			},
-			enc: "U291ci5pcwAAVNCqYhhnLPWrAAUBAgMBA7axWhhYd+HsAAUEBQYBAwICHr9ryhhdbkEZAAUHCAkBAwMCDy/UIhidCwCqAAUKCwwBA/NCwhh6wXgXAAUNDg8BAwUCHg",
-			out: [][]byte{{1, 2, 3}, {4, 5, 6}, {7, 8, 9}, {10, 11, 12}, {13, 14, 15}},
-			rev: [][]byte{{13, 14, 15}, {10, 11, 12}, {7, 8, 9}, {4, 5, 6}, {1, 2, 3}}},
-	}	
-
-	for _, test := range tests {
-		t.Run(test.name, func(t *testing.T) {
-			is := is.New(t)
-			buf := &buffer{}
-
-			buffers := 0
-
-			if len(test.in) == 0 {
-				err := WriteLogFile(buf, slices.Values([]io.Reader{}))
-				is.NoErr(err)
-			}
-			for i, in := range test.in {
-				buffers += len(in)
-
-				if i == 0 {
-					err := WriteLogFile(buf, slices.Values(in))
-					is.NoErr(err)
-				} else {
-					err := AppendLogFile(buf, slices.Values(in))
-					is.NoErr(err)
-				}
-			}
-
-			is.Equal(base64.RawStdEncoding.EncodeToString(buf.Bytes()), test.enc)
-
-			files, err := ReadLogFile(bytes.NewReader(buf.Bytes()))
-			is.NoErr(err)
-
-			i := 0
-			for j, fp := range files.Iter() {
-				buf, err := io.ReadAll(fp)
-				is.NoErr(err)
-
-				is.True(len(test.out) > int(j))
-				is.Equal(buf, test.out[j])
-				i++
-			}
-			is.NoErr(files.Err)
-			is.Equal(i, buffers)
-
-			i = 0
-			for j, fp := range files.Rev() {
-				buf, err := io.ReadAll(fp)
-				is.NoErr(err)
-
-				is.Equal(buf, test.rev[i])
-				is.Equal(buf, test.out[j])
-				i++
-			}
-			is.NoErr(files.Err)
-			is.Equal(i, buffers)
-
-		})
-	}
-}
-
-// TestArgs tests that the CLI arguments are correctly parsed.
-func TestArgs(t *testing.T) {
-	is := is.New(t)
-	usage := `Usage: lsm2 create <archive> <files>...`
-
-	arguments, err := docopt.ParseArgs(usage, []string{"create", "archive", "file1", "file2"}, "1.0")
-	is.NoErr(err)
-
-	var params struct {
-		Create  bool     `docopt:"create"`
-		Archive string   `docopt:"<archive>"`
-		Files   []string `docopt:"<files>"`
-	}
-	err = arguments.Bind(&params)
-	is.NoErr(err)
-
-	is.Equal(params.Create, true)
-	is.Equal(params.Archive, "archive")
-	is.Equal(params.Files, []string{"file1", "file2"})
-}
-
-type buffer struct {
-	buf []byte
-}
-
-// Bytes returns the underlying byte slice of the bufferWriterAt.
-func (b *buffer) Bytes() []byte {
-	return b.buf
-}
-
-// WriteAt implements io.WriterAt. It appends data to the internal buffer
-// if the offset is beyond the current length of the buffer. It will
-// return an error if the offset is negative.
-func (b *buffer) WriteAt(data []byte, offset int64) (written int, err error) {
-	if offset < 0 {
-		return 0, errors.New("negative offset")
-	}
-
-	currentLength := int64(len(b.buf))
-	if currentLength < offset+int64(len(data)) {
-		b.buf = append(b.buf, make([]byte, offset+int64(len(data))-currentLength)...)
-	}
-
-	written = copy(b.buf[offset:], data)
-	return
-}
-
-// ReadAt implements io.ReaderAt. It reads data from the internal buffer starting
-// from the specified offset and writes it into the provided data slice. If the
-// offset is negative, it returns an error. If the requested read extends beyond
-// the buffer's length, it returns the data read so far along with an io.EOF error.
-func (b *buffer) ReadAt(data []byte, offset int64) (int, error) {
-	if offset < 0 {
-		return 0, errors.New("negative offset")
-	}
-
-	if offset > int64(len(b.buf)) || len(b.buf[offset:]) < len(data) {
-		return copy(data, b.buf[offset:]), io.EOF
-	}
-
-	return copy(data, b.buf[offset:]), nil
-}
-
-// IterOne takes an iterator that yields values of type T along with a value of
-// type I, and returns an iterator that yields only the values of type T. It
-// discards the values of type I.
-func IterOne[I, T any](it iter.Seq2[I, T]) iter.Seq[T] {
-	return func(yield func(T) bool) {
-		for i, v := range it {
-			_ = i
-			if !yield(v) {
-				return
-			}
-		}
-	}
-}