config: support a configurable, and turn-off-able, pack.window

One use of go-git is to transfer git data from a non-standard git repo
(not stored in a file system, for example) to a "remote" backed by a
standard, local .git repo.

In this scenario, delta compression is not needed to reduce transfer
time over the "network", because there is no network. The underlying
storage layer has already taken care of the data tranfer, and sending
the objects to local .git storage doesn't require compression. So this
PR gives the user the option to turn off compression when it isn't
needed.

Of course, this results in a larger, uncompressed local .git repo, but
the user can then run git gc or git repack on that repo if they care
about the storage costs.

Turning the pack window to 0 on reduces total push time of a 36K repo
by 50 seconds (out of a pre-PR total of 3m26s).

Author: strib

config/config.go

@@ -6,6 +6,7 @@ import (
 	"errors"
 	"fmt"
 	"sort"
+	"strconv"
 
 	format "gopkg.in/src-d/go-git.v4/plumbing/format/config"
 )
@@ -40,6 +41,14 @@ type Config struct {
 		// Worktree is the path to the root of the working tree.
 		Worktree string
 	}
+
+	Pack struct {
+		// Window controls the size of the sliding window for delta
+		// compression.  The default is 10.  A value of 0 turns off
+		// delta compression entirely.
+		Window uint
+	}
+
 	// Remotes list of repository remotes, the key of the map is the name
 	// of the remote, should equal to RemoteConfig.Name.
 	Remotes map[string]*RemoteConfig
@@ -81,10 +90,14 @@ const (
 	remoteSection    = "remote"
 	submoduleSection = "submodule"
 	coreSection      = "core"
+	packSection      = "pack"
 	fetchKey         = "fetch"
 	urlKey           = "url"
 	bareKey          = "bare"
 	worktreeKey      = "worktree"
+	windowKey        = "window"
+
+	defaultPackWindow = uint(10)
 )
 
 // Unmarshal parses a git-config file and stores it.
@@ -98,6 +111,9 @@ func (c *Config) Unmarshal(b []byte) error {
 	}
 
 	c.unmarshalCore()
+	if err := c.unmarshalPack(); err != nil {
+		return err
+	}
 	c.unmarshalSubmodules()
 	return c.unmarshalRemotes()
 }
@@ -111,6 +127,21 @@ func (c *Config) unmarshalCore() {
 	c.Core.Worktree = s.Options.Get(worktreeKey)
 }
 
+func (c *Config) unmarshalPack() error {
+	s := c.Raw.Section(packSection)
+	window := s.Options.Get(windowKey)
+	if window == "" {
+		c.Pack.Window = defaultPackWindow
+	} else {
+		winUint, err := strconv.ParseUint(window, 10, 32)
+		if err != nil {
+			return err
+		}
+		c.Pack.Window = uint(winUint)
+	}
+	return nil
+}
+
 func (c *Config) unmarshalRemotes() error {
 	s := c.Raw.Section(remoteSection)
 	for _, sub := range s.Subsections {
@@ -138,6 +169,7 @@ func (c *Config) unmarshalSubmodules() {
 // Marshal returns Config encoded as a git-config file.
 func (c *Config) Marshal() ([]byte, error) {
 	c.marshalCore()
+	c.marshalPack()
 	c.marshalRemotes()
 	c.marshalSubmodules()
 
@@ -158,6 +190,13 @@ func (c *Config) marshalCore() {
 	}
 }
 
+func (c *Config) marshalPack() {
+	s := c.Raw.Section(packSection)
+	if c.Pack.Window != defaultPackWindow {
+		s.SetOption(windowKey, fmt.Sprintf("%d", c.Pack.Window))
+	}
+}
+
 func (c *Config) marshalRemotes() {
 	s := c.Raw.Section(remoteSection)
 	newSubsections := make(format.Subsections, 0, len(c.Remotes))

config/config_test.go

@@ -10,6 +10,8 @@ func (s *ConfigSuite) TestUnmarshall(c *C) {
 	input := []byte(`[core]
         bare = true
 		worktree = foo
+[pack]
+		window = 20
 [remote "origin"]
         url = [email protected]:mcuadros/go-git.git
         fetch = +refs/heads/*:refs/remotes/origin/*
@@ -33,6 +35,7 @@ func (s *ConfigSuite) TestUnmarshall(c *C) {
 
 	c.Assert(cfg.Core.IsBare, Equals, true)
 	c.Assert(cfg.Core.Worktree, Equals, "foo")
+	c.Assert(cfg.Pack.Window, Equals, uint(20))
 	c.Assert(cfg.Remotes, HasLen, 2)
 	c.Assert(cfg.Remotes["origin"].Name, Equals, "origin")
 	c.Assert(cfg.Remotes["origin"].URLs, DeepEquals, []string{"[email protected]:mcuadros/go-git.git"})
@@ -51,6 +54,8 @@ func (s *ConfigSuite) TestMarshall(c *C) {
 	output := []byte(`[core]
 	bare = true
 	worktree = bar
+[pack]
+	window = 20
 [remote "alt"]
 	url = [email protected]:mcuadros/go-git.git
 	url = [email protected]:src-d/go-git.git
@@ -65,6 +70,7 @@ func (s *ConfigSuite) TestMarshall(c *C) {
 	cfg := NewConfig()
 	cfg.Core.IsBare = true
 	cfg.Core.Worktree = "bar"
+	cfg.Pack.Window = 20
 	cfg.Remotes["origin"] = &RemoteConfig{
 		Name: "origin",
 		URLs: []string{"[email protected]:mcuadros/go-git.git"},
@@ -92,6 +98,8 @@ func (s *ConfigSuite) TestUnmarshallMarshall(c *C) {
 	bare = true
 	worktree = foo
 	custom = ignored
+[pack]
+	window = 20
 [remote "origin"]
 	url = [email protected]:mcuadros/go-git.git
 	fetch = +refs/heads/*:refs/remotes/origin/*

plumbing/format/packfile/delta_selector.go

@@ -9,9 +9,6 @@ import (
 )
 
 const (
-	// How far back in the sorted list to search for deltas.  10 is
-	// the default in command line git.
-	deltaWindowSize = 10
 	// deltas based on deltas, how many steps we can do.
 	// 50 is the default value used in JGit
 	maxDepth = int64(50)
@@ -33,12 +30,19 @@ func newDeltaSelector(s storer.EncodedObjectStorer) *deltaSelector {
 
 // ObjectsToPack creates a list of ObjectToPack from the hashes provided,
 // creating deltas if it's suitable, using an specific internal logic
-func (dw *deltaSelector) ObjectsToPack(hashes []plumbing.Hash) ([]*ObjectToPack, error) {
-	otp, err := dw.objectsToPack(hashes)
+func (dw *deltaSelector) ObjectsToPack(
+	hashes []plumbing.Hash,
+	packWindow uint,
+) ([]*ObjectToPack, error) {
+	otp, err := dw.objectsToPack(hashes, packWindow)
 	if err != nil {
 		return nil, err
 	}
 
+	if packWindow == 0 {
+		return otp, nil
+	}
+
 	dw.sort(otp)
 
 	var objectGroups [][]*ObjectToPack
@@ -60,7 +64,7 @@ func (dw *deltaSelector) ObjectsToPack(hashes []plumbing.Hash) ([]*ObjectToPack,
 		objs := objs
 		wg.Add(1)
 		go func() {
-			if walkErr := dw.walk(objs); walkErr != nil {
+			if walkErr := dw.walk(objs, packWindow); walkErr != nil {
 				once.Do(func() {
 					err = walkErr
 				})
@@ -77,10 +81,19 @@ func (dw *deltaSelector) ObjectsToPack(hashes []plumbing.Hash) ([]*ObjectToPack,
 	return otp, nil
 }
 
-func (dw *deltaSelector) objectsToPack(hashes []plumbing.Hash) ([]*ObjectToPack, error) {
+func (dw *deltaSelector) objectsToPack(
+	hashes []plumbing.Hash,
+	packWindow uint,
+) ([]*ObjectToPack, error) {
 	var objectsToPack []*ObjectToPack
 	for _, h := range hashes {
-		o, err := dw.encodedDeltaObject(h)
+		var o plumbing.EncodedObject
+		var err error
+		if packWindow == 0 {
+			o, err = dw.encodedObject(h)
+		} else {
+			o, err = dw.encodedDeltaObject(h)
+		}
 		if err != nil {
 			return nil, err
 		}
@@ -93,6 +106,10 @@ func (dw *deltaSelector) objectsToPack(hashes []plumbing.Hash) ([]*ObjectToPack,
 		objectsToPack = append(objectsToPack, otp)
 	}
 
+	if packWindow == 0 {
+		return objectsToPack, nil
+	}
+
 	if err := dw.fixAndBreakChains(objectsToPack); err != nil {
 		return nil, err
 	}
@@ -201,7 +218,10 @@ func (dw *deltaSelector) sort(objectsToPack []*ObjectToPack) {
 	sort.Sort(byTypeAndSize(objectsToPack))
 }
 
-func (dw *deltaSelector) walk(objectsToPack []*ObjectToPack) error {
+func (dw *deltaSelector) walk(
+	objectsToPack []*ObjectToPack,
+	packWindow uint,
+) error {
 	indexMap := make(map[plumbing.Hash]*deltaIndex)
 	for i := 0; i < len(objectsToPack); i++ {
 		target := objectsToPack[i]
@@ -218,7 +238,7 @@ func (dw *deltaSelector) walk(objectsToPack []*ObjectToPack) error {
 			continue
 		}
 
-		for j := i - 1; j >= 0 && i-j < deltaWindowSize; j-- {
+		for j := i - 1; j >= 0 && i-j < int(packWindow); j-- {
 			base := objectsToPack[j]
 			// Objects must use only the same type as their delta base.
 			// Since objectsToPack is sorted by type and size, once we find

plumbing/format/packfile/delta_selector_test.go

@@ -146,31 +146,32 @@ func (s *DeltaSelectorSuite) createTestObjects() {
 func (s *DeltaSelectorSuite) TestObjectsToPack(c *C) {
 	// Different type
 	hashes := []plumbing.Hash{s.hashes["base"], s.hashes["treeType"]}
-	otp, err := s.ds.ObjectsToPack(hashes)
+	deltaWindowSize := uint(10)
+	otp, err := s.ds.ObjectsToPack(hashes, deltaWindowSize)
 	c.Assert(err, IsNil)
 	c.Assert(len(otp), Equals, 2)
 	c.Assert(otp[0].Object, Equals, s.store.Objects[s.hashes["base"]])
 	c.Assert(otp[1].Object, Equals, s.store.Objects[s.hashes["treeType"]])
 
 	// Size radically different
 	hashes = []plumbing.Hash{s.hashes["bigBase"], s.hashes["target"]}
-	otp, err = s.ds.ObjectsToPack(hashes)
+	otp, err = s.ds.ObjectsToPack(hashes, deltaWindowSize)
 	c.Assert(err, IsNil)
 	c.Assert(len(otp), Equals, 2)
 	c.Assert(otp[0].Object, Equals, s.store.Objects[s.hashes["bigBase"]])
 	c.Assert(otp[1].Object, Equals, s.store.Objects[s.hashes["target"]])
 
 	// Delta Size Limit with no best delta yet
 	hashes = []plumbing.Hash{s.hashes["smallBase"], s.hashes["smallTarget"]}
-	otp, err = s.ds.ObjectsToPack(hashes)
+	otp, err = s.ds.ObjectsToPack(hashes, deltaWindowSize)
 	c.Assert(err, IsNil)
 	c.Assert(len(otp), Equals, 2)
 	c.Assert(otp[0].Object, Equals, s.store.Objects[s.hashes["smallBase"]])
 	c.Assert(otp[1].Object, Equals, s.store.Objects[s.hashes["smallTarget"]])
 
 	// It will create the delta
 	hashes = []plumbing.Hash{s.hashes["base"], s.hashes["target"]}
-	otp, err = s.ds.ObjectsToPack(hashes)
+	otp, err = s.ds.ObjectsToPack(hashes, deltaWindowSize)
 	c.Assert(err, IsNil)
 	c.Assert(len(otp), Equals, 2)
 	c.Assert(otp[0].Object, Equals, s.store.Objects[s.hashes["target"]])
@@ -185,7 +186,7 @@ func (s *DeltaSelectorSuite) TestObjectsToPack(c *C) {
 		s.hashes["o2"],
 		s.hashes["o3"],
 	}
-	otp, err = s.ds.ObjectsToPack(hashes)
+	otp, err = s.ds.ObjectsToPack(hashes, deltaWindowSize)
 	c.Assert(err, IsNil)
 	c.Assert(len(otp), Equals, 3)
 	c.Assert(otp[0].Object, Equals, s.store.Objects[s.hashes["o1"]])
@@ -201,20 +202,32 @@ func (s *DeltaSelectorSuite) TestObjectsToPack(c *C) {
 	// a delta.
 	hashes = make([]plumbing.Hash, 0, deltaWindowSize+2)
 	hashes = append(hashes, s.hashes["base"])
-	for i := 0; i < deltaWindowSize; i++ {
+	for i := uint(0); i < deltaWindowSize; i++ {
 		hashes = append(hashes, s.hashes["smallTarget"])
 	}
 	hashes = append(hashes, s.hashes["target"])
 
 	// Don't sort so we can easily check the sliding window without
 	// creating a bunch of new objects.
-	otp, err = s.ds.objectsToPack(hashes)
+	otp, err = s.ds.objectsToPack(hashes, deltaWindowSize)
 	c.Assert(err, IsNil)
-	err = s.ds.walk(otp)
+	err = s.ds.walk(otp, deltaWindowSize)
 	c.Assert(err, IsNil)
-	c.Assert(len(otp), Equals, deltaWindowSize+2)
+	c.Assert(len(otp), Equals, int(deltaWindowSize)+2)
 	targetIdx := len(otp) - 1
 	c.Assert(otp[targetIdx].IsDelta(), Equals, false)
+
+	// Check that no deltas are created, and the objects are unsorted,
+	// if compression is off.
+	hashes = []plumbing.Hash{s.hashes["base"], s.hashes["target"]}
+	otp, err = s.ds.ObjectsToPack(hashes, 0)
+	c.Assert(err, IsNil)
+	c.Assert(len(otp), Equals, 2)
+	c.Assert(otp[0].Object, Equals, s.store.Objects[s.hashes["base"]])
+	c.Assert(otp[0].IsDelta(), Equals, false)
+	c.Assert(otp[1].Original, Equals, s.store.Objects[s.hashes["target"]])
+	c.Assert(otp[1].IsDelta(), Equals, false)
+	c.Assert(otp[1].Depth, Equals, 0)
 }
 
 func (s *DeltaSelectorSuite) TestMaxDepth(c *C) {

plumbing/format/packfile/encoder.go

@@ -14,10 +14,10 @@ import (
 // Encoder gets the data from the storage and write it into the writer in PACK
 // format
 type Encoder struct {
-	selector     *deltaSelector
-	w            *offsetWriter
-	zw           *zlib.Writer
-	hasher       plumbing.Hasher
+	selector *deltaSelector
+	w        *offsetWriter
+	zw       *zlib.Writer
+	hasher   plumbing.Hasher
 	// offsets is a map of object hashes to corresponding offsets in the packfile.
 	// It is used to determine offset of the base of a delta when a OFS_DELTA is
 	// used.
@@ -47,8 +47,11 @@ func NewEncoder(w io.Writer, s storer.EncodedObjectStorer, useRefDeltas bool) *E
 
 // Encode creates a packfile containing all the objects referenced in hashes
 // and writes it to the writer in the Encoder.
-func (e *Encoder) Encode(hashes []plumbing.Hash) (plumbing.Hash, error) {
-	objects, err := e.selector.ObjectsToPack(hashes)
+func (e *Encoder) Encode(
+	hashes []plumbing.Hash,
+	packWindow uint,
+) (plumbing.Hash, error) {
+	objects, err := e.selector.ObjectsToPack(hashes, packWindow)
 	if err != nil {
 		return plumbing.ZeroHash, err
 	}
@@ -137,7 +140,7 @@ func (e *Encoder) writeOfsDeltaHeader(deltaOffset int64, base plumbing.Hash) err
 
 	// for OFS_DELTA, offset of the base is interpreted as negative offset
 	// relative to the type-byte of the header of the ofs-delta entry.
-	relativeOffset := deltaOffset-baseOffset
+	relativeOffset := deltaOffset - baseOffset
 	if relativeOffset <= 0 {
 		return fmt.Errorf("bad offset for OFS_DELTA entry: %d", relativeOffset)
 	}

plumbing/format/packfile/encoder_advanced_test.go

@@ -27,12 +27,23 @@ func (s *EncoderAdvancedSuite) TestEncodeDecode(c *C) {
 	fixs.Test(c, func(f *fixtures.Fixture) {
 		storage, err := filesystem.NewStorage(f.DotGit())
 		c.Assert(err, IsNil)
-		s.testEncodeDecode(c, storage)
+		s.testEncodeDecode(c, storage, 10)
 	})
 
 }
 
-func (s *EncoderAdvancedSuite) testEncodeDecode(c *C, storage storer.Storer) {
+func (s *EncoderAdvancedSuite) TestEncodeDecodeNoDeltaCompression(c *C) {
+	fixs := fixtures.Basic().ByTag("packfile").ByTag(".git")
+	fixs = append(fixs, fixtures.ByURL("https://github.com/src-d/go-git.git").
+		ByTag("packfile").ByTag(".git").One())
+	fixs.Test(c, func(f *fixtures.Fixture) {
+		storage, err := filesystem.NewStorage(f.DotGit())
+		c.Assert(err, IsNil)
+		s.testEncodeDecode(c, storage, 0)
+	})
+}
+
+func (s *EncoderAdvancedSuite) testEncodeDecode(c *C, storage storer.Storer, packWindow uint) {
 
 	objIter, err := storage.IterEncodedObjects(plumbing.AnyObject)
 	c.Assert(err, IsNil)
@@ -57,7 +68,7 @@ func (s *EncoderAdvancedSuite) testEncodeDecode(c *C, storage storer.Storer) {
 
 	buf := bytes.NewBuffer(nil)
 	enc := NewEncoder(buf, storage, false)
-	_, err = enc.Encode(hashes)
+	_, err = enc.Encode(hashes, packWindow)
 	c.Assert(err, IsNil)
 
 	scanner := NewScanner(buf)