diff options
38 files changed, 1547 insertions, 1516 deletions
@@ -426,7 +426,7 @@ func (bug *Bug) Commit(repo repository.ClockedRepo) error { // Write a Git commit referencing the tree, with the previous commit as parent if bug.lastCommit != "" { - hash, err = repo.StoreCommitWithParent(hash, bug.lastCommit) + hash, err = repo.StoreCommit(hash, bug.lastCommit) } else { hash, err = repo.StoreCommit(hash) } @@ -524,7 +524,7 @@ func (bug *Bug) Merge(repo repository.Repo, other Interface) (bool, error) { } // create a new commit with the correct ancestor - hash, err := repo.StoreCommitWithParent(treeHash, bug.lastCommit) + hash, err := repo.StoreCommit(treeHash, bug.lastCommit) if err != nil { return false, err diff --git a/entity/TODO b/entity/TODO new file mode 100644 index 00000000..fd3c9710 --- /dev/null +++ b/entity/TODO @@ -0,0 +1,8 @@ +- is the pack Lamport clock really useful vs only topological sort? + - topological order is enforced on the clocks, so what's the point? + - is EditTime equivalent to PackTime? no, avoid the gaps. Is it better? +- how to do commit signature? +- how to avoid id collision between Operations? +- write tests for actions +- migrate Bug to the new structure +- migrate Identity to the new structure?
\ No newline at end of file diff --git a/entity/dag/common_test.go b/entity/dag/common_test.go new file mode 100644 index 00000000..29f1279e --- /dev/null +++ b/entity/dag/common_test.go @@ -0,0 +1,137 @@ +package dag + +import ( + "encoding/json" + "fmt" + + "github.com/MichaelMure/git-bug/entity" + "github.com/MichaelMure/git-bug/identity" + "github.com/MichaelMure/git-bug/repository" +) + +// This file contains an example dummy entity to be used in the tests + +/* + Operations +*/ + +type op1 struct { + author identity.Interface + + OperationType int `json:"type"` + Field1 string `json:"field_1"` +} + +func newOp1(author identity.Interface, field1 string) *op1 { + return &op1{author: author, OperationType: 1, Field1: field1} +} + +func (o op1) Id() entity.Id { + data, _ := json.Marshal(o) + return entity.DeriveId(data) +} + +func (o op1) Author() identity.Interface { + return o.author +} + +func (o op1) Validate() error { return nil } + +type op2 struct { + author identity.Interface + + OperationType int `json:"type"` + Field2 string `json:"field_2"` +} + +func newOp2(author identity.Interface, field2 string) *op2 { + return &op2{author: author, OperationType: 2, Field2: field2} +} + +func (o op2) Id() entity.Id { + data, _ := json.Marshal(o) + return entity.DeriveId(data) +} + +func (o op2) Author() identity.Interface { + return o.author +} + +func (o op2) Validate() error { return nil } + +func unmarshaler(author identity.Interface, raw json.RawMessage) (Operation, error) { + var t struct { + OperationType int `json:"type"` + } + + if err := json.Unmarshal(raw, &t); err != nil { + return nil, err + } + + switch t.OperationType { + case 1: + op := &op1{} + err := json.Unmarshal(raw, &op) + op.author = author + return op, err + case 2: + op := &op2{} + err := json.Unmarshal(raw, &op) + op.author = author + return op, err + default: + return nil, fmt.Errorf("unknown operation type %v", t.OperationType) + } +} + +/* + Identities + repo + definition +*/ + +func makeTestContext() (repository.ClockedRepo, identity.Interface, identity.Interface, Definition) { + repo := repository.NewMockRepo() + + id1, err := identity.NewIdentity(repo, "name1", "email1") + if err != nil { + panic(err) + } + err = id1.Commit(repo) + if err != nil { + panic(err) + } + id2, err := identity.NewIdentity(repo, "name2", "email2") + if err != nil { + panic(err) + } + err = id2.Commit(repo) + if err != nil { + panic(err) + } + + resolver := identityResolverFunc(func(id entity.Id) (identity.Interface, error) { + switch id { + case id1.Id(): + return id1, nil + case id2.Id(): + return id2, nil + default: + return nil, identity.ErrIdentityNotExist + } + }) + + def := Definition{ + typename: "foo", + namespace: "foos", + operationUnmarshaler: unmarshaler, + identityResolver: resolver, + formatVersion: 1, + } + + return repo, id1, id2, def +} + +type identityResolverFunc func(id entity.Id) (identity.Interface, error) + +func (fn identityResolverFunc) ResolveIdentity(id entity.Id) (identity.Interface, error) { + return fn(id) +} diff --git a/entity/entity.go b/entity/dag/entity.go index a1e8e57e..78347fa0 100644 --- a/entity/entity.go +++ b/entity/dag/entity.go @@ -1,4 +1,6 @@ -package entity +// Package dag contains the base common code to define an entity stored +// in a chain of git objects, supporting actions like Push, Pull and Merge. +package dag import ( "encoding/json" @@ -7,6 +9,8 @@ import ( "github.com/pkg/errors" + "github.com/MichaelMure/git-bug/entity" + "github.com/MichaelMure/git-bug/identity" "github.com/MichaelMure/git-bug/repository" "github.com/MichaelMure/git-bug/util/lamport" ) @@ -15,12 +19,6 @@ const refsPattern = "refs/%s/%s" const creationClockPattern = "%s-create" const editClockPattern = "%s-edit" -type Operation interface { - Id() Id - // MarshalJSON() ([]byte, error) - Validate() error -} - // Definition hold the details defining one specialization of an Entity. type Definition struct { // the name of the entity (bug, pull-request, ...) @@ -28,29 +26,40 @@ type Definition struct { // the namespace in git (bugs, prs, ...) namespace string // a function decoding a JSON message into an Operation - operationUnmarshaler func(raw json.RawMessage) (Operation, error) - // the expected format version number + operationUnmarshaler func(author identity.Interface, raw json.RawMessage) (Operation, error) + // a function loading an identity.Identity from its Id + identityResolver identity.Resolver + // the expected format version number, that can be used for data migration/upgrade formatVersion uint } +// Entity is a data structure stored in a chain of git objects, supporting actions like Push, Pull and Merge. type Entity struct { Definition - ops []Operation + // operations that are already stored in the repository + ops []Operation + // operations not yet stored in the repository staging []Operation - packClock lamport.Clock + // TODO: add here createTime and editTime + + // // TODO: doesn't seems to actually be useful over the topological sort ? Timestamp can be generated from graph depth + // // TODO: maybe EditTime is better because it could spread ops in consecutive groups on the logical timeline --> avoid interleaving + // packClock lamport.Clock lastCommit repository.Hash } +// New create an empty Entity func New(definition Definition) *Entity { return &Entity{ Definition: definition, - packClock: lamport.NewMemClock(), + // packClock: lamport.NewMemClock(), } } -func Read(def Definition, repo repository.ClockedRepo, id Id) (*Entity, error) { +// Read will read and decode a stored Entity from a repository +func Read(def Definition, repo repository.ClockedRepo, id entity.Id) (*Entity, error) { if err := id.Validate(); err != nil { return nil, errors.Wrap(err, "invalid id") } @@ -109,32 +118,33 @@ func read(def Definition, repo repository.ClockedRepo, ref string) (*Entity, err oppMap := make(map[repository.Hash]*operationPack) var opsCount int - var packClock = lamport.NewMemClock() + // var packClock = lamport.NewMemClock() for i := len(DFSOrder) - 1; i >= 0; i-- { commit := DFSOrder[i] - firstCommit := i == len(DFSOrder)-1 + isFirstCommit := i == len(DFSOrder)-1 + isMerge := len(commit.Parents) > 1 // Verify DAG structure: single chronological root, so only the root - // can have no parents - if !firstCommit && len(commit.Parents) == 0 { - return nil, fmt.Errorf("multiple root in the entity DAG") + // can have no parents. Said otherwise, the DAG need to have exactly + // one leaf. + if !isFirstCommit && len(commit.Parents) == 0 { + return nil, fmt.Errorf("multiple leafs in the entity DAG") } - opp, err := readOperationPack(def, repo, commit.TreeHash) + opp, err := readOperationPack(def, repo, commit) if err != nil { return nil, err } - // Check that the lamport clocks are set - if firstCommit && opp.CreateTime <= 0 { - return nil, fmt.Errorf("creation lamport time not set") - } - if opp.EditTime <= 0 { - return nil, fmt.Errorf("edition lamport time not set") + err = opp.Validate() + if err != nil { + return nil, err } - if opp.PackTime <= 0 { - return nil, fmt.Errorf("pack lamport time not set") + + // Check that the create lamport clock is set (not checked in Validate() as it's optional) + if isFirstCommit && opp.CreateTime <= 0 { + return nil, fmt.Errorf("creation lamport time not set") } // make sure that the lamport clocks causality match the DAG topology @@ -150,9 +160,13 @@ func read(def Definition, repo repository.ClockedRepo, ref string) (*Entity, err // to avoid an attack where clocks are pushed toward the uint64 rollover, make sure // that the clocks don't jump too far in the future - if opp.EditTime-parentPack.EditTime > 10_000 { + // we ignore merge commits here to allow merging after a loooong time without breaking anything, + // as long as there is one valid chain of small hops, it's fine. + if !isMerge && opp.EditTime-parentPack.EditTime > 1_000_000 { return nil, fmt.Errorf("lamport clock jumping too far in the future, likely an attack") } + + // TODO: PackTime is not checked } oppMap[commit.Hash] = opp @@ -169,10 +183,10 @@ func read(def Definition, repo repository.ClockedRepo, ref string) (*Entity, err if err != nil { return nil, err } - err = packClock.Witness(opp.PackTime) - if err != nil { - return nil, err - } + // err = packClock.Witness(opp.PackTime) + // if err != nil { + // return nil, err + // } } // Now that we know that the topological order and clocks are fine, we order the operationPacks @@ -185,20 +199,20 @@ func read(def Definition, repo repository.ClockedRepo, ref string) (*Entity, err sort.Slice(oppSlice, func(i, j int) bool { // Primary ordering with the dedicated "pack" Lamport time that encode causality // within the entity - if oppSlice[i].PackTime != oppSlice[j].PackTime { - return oppSlice[i].PackTime < oppSlice[i].PackTime - } + // if oppSlice[i].PackTime != oppSlice[j].PackTime { + // return oppSlice[i].PackTime < oppSlice[i].PackTime + // } // We have equal PackTime, which means we had a concurrent edition. We can't tell which exactly // came first. As a secondary arbitrary ordering, we can use the EditTime. It's unlikely to be // enough but it can give us an edge to approach what really happened. if oppSlice[i].EditTime != oppSlice[j].EditTime { return oppSlice[i].EditTime < oppSlice[j].EditTime } - // Well, what now? We still need a total ordering, the most stable possible. + // Well, what now? We still need a total ordering and the most stable possible. // As a last resort, we can order based on a hash of the serialized Operations in the // operationPack. It doesn't carry much meaning but it's unbiased and hard to abuse. - // This is a lexicographic ordering. - return oppSlice[i].Id < oppSlice[j].Id + // This is a lexicographic ordering on the stringified ID. + return oppSlice[i].Id() < oppSlice[j].Id() }) // Now that we ordered the operationPacks, we have the order of the Operations @@ -213,16 +227,18 @@ func read(def Definition, repo repository.ClockedRepo, ref string) (*Entity, err return &Entity{ Definition: def, ops: ops, + // packClock: packClock, lastCommit: rootHash, }, nil } // Id return the Entity identifier -func (e *Entity) Id() Id { +func (e *Entity) Id() entity.Id { // id is the id of the first operation return e.FirstOp().Id() } +// Validate check if the Entity data is valid func (e *Entity) Validate() error { // non-empty if len(e.ops) == 0 && len(e.staging) == 0 { @@ -244,7 +260,7 @@ func (e *Entity) Validate() error { } // Check that there is no colliding operation's ID - ids := make(map[Id]struct{}) + ids := make(map[entity.Id]struct{}) for _, op := range e.Operations() { if _, ok := ids[op.Id()]; ok { return fmt.Errorf("id collision: %s", op.Id()) @@ -255,12 +271,12 @@ func (e *Entity) Validate() error { return nil } -// return the ordered operations +// Operations return the ordered operations func (e *Entity) Operations() []Operation { return append(e.ops, e.staging...) } -// Lookup for the very first operation of the Entity. +// FirstOp lookup for the very first operation of the Entity func (e *Entity) FirstOp() Operation { for _, op := range e.ops { return op @@ -271,14 +287,29 @@ func (e *Entity) FirstOp() Operation { return nil } +// LastOp lookup for the very last operation of the Entity +func (e *Entity) LastOp() Operation { + if len(e.staging) > 0 { + return e.staging[len(e.staging)-1] + } + if len(e.ops) > 0 { + return e.ops[len(e.ops)-1] + } + return nil +} + +// Append add a new Operation to the Entity func (e *Entity) Append(op Operation) { e.staging = append(e.staging, op) } +// NeedCommit indicate if the in-memory state changed and need to be commit in the repository func (e *Entity) NeedCommit() bool { return len(e.staging) > 0 } +// CommitAdNeeded execute a Commit only if necessary. This function is useful to avoid getting an error if the Entity +// is already in sync with the repository. func (e *Entity) CommitAdNeeded(repo repository.ClockedRepo) error { if e.NeedCommit() { return e.Commit(repo) @@ -286,6 +317,7 @@ func (e *Entity) CommitAdNeeded(repo repository.ClockedRepo) error { return nil } +// Commit write the appended operations in the repository // TODO: support commit signature func (e *Entity) Commit(repo repository.ClockedRepo) error { if !e.NeedCommit() { @@ -296,11 +328,19 @@ func (e *Entity) Commit(repo repository.ClockedRepo) error { return errors.Wrapf(err, "can't commit a %s with invalid data", e.Definition.typename) } - // increment the various clocks for this new operationPack - packTime, err := e.packClock.Increment() - if err != nil { - return err + var author identity.Interface + for _, op := range e.staging { + if author != nil && op.Author() != author { + return fmt.Errorf("operations with different author") + } + author = op.Author() } + + // increment the various clocks for this new operationPack + // packTime, err := e.packClock.Increment() + // if err != nil { + // return err + // } editTime, err := repo.Increment(fmt.Sprintf(editClockPattern, e.namespace)) if err != nil { return err @@ -314,13 +354,14 @@ func (e *Entity) Commit(repo repository.ClockedRepo) error { } opp := &operationPack{ + Author: author, Operations: e.staging, CreateTime: creationTime, EditTime: editTime, - PackTime: packTime, + // PackTime: packTime, } - treeHash, err := opp.write(e.Definition, repo) + treeHash, err := opp.Write(e.Definition, repo) if err != nil { return err } @@ -328,7 +369,7 @@ func (e *Entity) Commit(repo repository.ClockedRepo) error { // Write a Git commit referencing the tree, with the previous commit as parent var commitHash repository.Hash if e.lastCommit != "" { - commitHash, err = repo.StoreCommitWithParent(treeHash, e.lastCommit) + commitHash, err = repo.StoreCommit(treeHash, e.lastCommit) } else { commitHash, err = repo.StoreCommit(treeHash) } diff --git a/entity/dag/entity_actions.go b/entity/dag/entity_actions.go new file mode 100644 index 00000000..8dcf91e6 --- /dev/null +++ b/entity/dag/entity_actions.go @@ -0,0 +1,227 @@ +package dag + +import ( + "fmt" + + "github.com/pkg/errors" + + "github.com/MichaelMure/git-bug/entity" + "github.com/MichaelMure/git-bug/repository" +) + +func ListLocalIds(typename string, repo repository.RepoData) ([]entity.Id, error) { + refs, err := repo.ListRefs(fmt.Sprintf("refs/%s/", typename)) + if err != nil { + return nil, err + } + return entity.RefsToIds(refs), nil +} + +// Fetch retrieve updates from a remote +// This does not change the local entity state +func Fetch(def Definition, repo repository.Repo, remote string) (string, error) { + // "refs/<entity>/*:refs/remotes/<remote>/<entity>/*" + fetchRefSpec := fmt.Sprintf("refs/%s/*:refs/remotes/%s/%s/*", + def.namespace, remote, def.namespace) + + return repo.FetchRefs(remote, fetchRefSpec) +} + +// Push update a remote with the local changes +func Push(def Definition, repo repository.Repo, remote string) (string, error) { + // "refs/<entity>/*:refs/<entity>/*" + refspec := fmt.Sprintf("refs/%s/*:refs/%s/*", + def.namespace, def.namespace) + + return repo.PushRefs(remote, refspec) +} + +// Pull will do a Fetch + MergeAll +// Contrary to MergeAll, this function will return an error if a merge fail. +func Pull(def Definition, repo repository.ClockedRepo, remote string) error { + _, err := Fetch(def, repo, remote) + if err != nil { + return err + } + + for merge := range MergeAll(def, repo, remote) { + if merge.Err != nil { + return merge.Err + } + if merge.Status == entity.MergeStatusInvalid { + return errors.Errorf("merge failure: %s", merge.Reason) + } + } + + return nil +} + +func MergeAll(def Definition, repo repository.ClockedRepo, remote string) <-chan entity.MergeResult { + out := make(chan entity.MergeResult) + + // no caching for the merge, we load everything from git even if that means multiple + // copy of the same entity in memory. The cache layer will intercept the results to + // invalidate entities if necessary. + + go func() { + defer close(out) + + remoteRefSpec := fmt.Sprintf("refs/remotes/%s/%s/", remote, def.namespace) + remoteRefs, err := repo.ListRefs(remoteRefSpec) + if err != nil { + out <- entity.MergeResult{Err: err} + return + } + + for _, remoteRef := range remoteRefs { + out <- merge(def, repo, remoteRef) + } + }() + + return out +} + +func merge(def Definition, repo repository.ClockedRepo, remoteRef string) entity.MergeResult { + id := entity.RefToId(remoteRef) + + if err := id.Validate(); err != nil { + return entity.NewMergeInvalidStatus(id, errors.Wrap(err, "invalid ref").Error()) + } + + remoteEntity, err := read(def, repo, remoteRef) + if err != nil { + return entity.NewMergeInvalidStatus(id, + errors.Wrapf(err, "remote %s is not readable", def.typename).Error()) + } + + // Check for error in remote data + if err := remoteEntity.Validate(); err != nil { + return entity.NewMergeInvalidStatus(id, + errors.Wrapf(err, "remote %s data is invalid", def.typename).Error()) + } + + localRef := fmt.Sprintf("refs/%s/%s", def.namespace, id.String()) + + localExist, err := repo.RefExist(localRef) + if err != nil { + return entity.NewMergeError(err, id) + } + + // the bug is not local yet, simply create the reference + if !localExist { + err := repo.CopyRef(remoteRef, localRef) + if err != nil { + return entity.NewMergeError(err, id) + } + + return entity.NewMergeStatus(entity.MergeStatusNew, id, remoteEntity) + } + + // var updated bool + // err = repo.MergeRef(localRef, remoteRef, func() repository.Hash { + // updated = true + // + // }) + // if err != nil { + // return entity.NewMergeError(err, id) + // } + // + // if updated { + // return entity.NewMergeStatus(entity.MergeStatusUpdated, id, ) + // } else { + // return entity.NewMergeStatus(entity.MergeStatusNothing, id, ) + // } + + localCommit, err := repo.ResolveRef(localRef) + if err != nil { + return entity.NewMergeError(err, id) + } + + remoteCommit, err := repo.ResolveRef(remoteRef) + if err != nil { + return entity.NewMergeError(err, id) + } + + if localCommit == remoteCommit { + // nothing to merge + return entity.NewMergeStatus(entity.MergeStatusNothing, id, remoteEntity) + } + + // fast-forward is possible if otherRef include ref + + remoteCommits, err := repo.ListCommits(remoteRef) + if err != nil { + return entity.NewMergeError(err, id) + } + + fastForwardPossible := false + for _, hash := range remoteCommits { + if hash == localCommit { + fastForwardPossible = true + break + } + } + + if fastForwardPossible { + err = repo.UpdateRef(localRef, remoteCommit) + if err != nil { + return entity.NewMergeError(err, id) + } + return entity.NewMergeStatus(entity.MergeStatusUpdated, id, remoteEntity) + } + + // fast-forward is not possible, we need to create a merge commit + // For simplicity when reading and to have clocks that record this change, we store + // an empty operationPack. + // First step is to collect those clocks. + + localEntity, err := read(def, repo, localRef) + if err != nil { + return entity.NewMergeError(err, id) + } + + // err = localEntity.packClock.Witness(remoteEntity.packClock.Time()) + // if err != nil { + // return entity.NewMergeError(err, id) + // } + // + // packTime, err := localEntity.packClock.Increment() + // if err != nil { + // return entity.NewMergeError(err, id) + // } + + editTime, err := repo.Increment(fmt.Sprintf(editClockPattern, def.namespace)) + if err != nil { + return entity.NewMergeError(err, id) + } + + opp := &operationPack{ + Operations: nil, + CreateTime: 0, + EditTime: editTime, + // PackTime: packTime, + } + + treeHash, err := opp.Write(def, repo) + if err != nil { + return entity.NewMergeError(err, id) + } + + // Create the merge commit with two parents + newHash, err := repo.StoreCommit(treeHash, localCommit, remoteCommit) + if err != nil { + return entity.NewMergeError(err, id) + } + + // finally update the ref + err = repo.UpdateRef(localRef, newHash) + if err != nil { + return entity.NewMergeError(err, id) + } + + return entity.NewMergeStatus(entity.MergeStatusUpdated, id, localEntity) +} + +func Remove() error { + panic("") +} diff --git a/entity/dag/entity_test.go b/entity/dag/entity_test.go new file mode 100644 index 00000000..c5c83567 --- /dev/null +++ b/entity/dag/entity_test.go @@ -0,0 +1,117 @@ +package dag + +import ( + "testing" + + "github.com/stretchr/testify/require" +) + +func TestWriteRead(t *testing.T) { + repo, id1, id2, def := makeTestContext() + + entity := New(def) + require.False(t, entity.NeedCommit()) + + entity.Append(newOp1(id1, "foo")) + entity.Append(newOp2(id1, "bar")) + + require.True(t, entity.NeedCommit()) + require.NoError(t, entity.CommitAdNeeded(repo)) + require.False(t, entity.NeedCommit()) + + entity.Append(newOp2(id2, "foobar")) + require.True(t, entity.NeedCommit()) + require.NoError(t, entity.CommitAdNeeded(repo)) + require.False(t, entity.NeedCommit()) + + read, err := Read(def, repo, entity.Id()) + require.NoError(t, err) + + assertEqualEntities(t, entity, read) +} + +func assertEqualEntities(t *testing.T, a, b *Entity) { + // testify doesn't support comparing functions and systematically fail if they are not nil + // so we have to set them to nil temporarily + + backOpUnA := a.Definition.operationUnmarshaler + backOpUnB := b.Definition.operationUnmarshaler + + a.Definition.operationUnmarshaler = nil + b.Definition.operationUnmarshaler = nil + + backIdResA := a.Definition.identityResolver + backIdResB := b.Definition.identityResolver + + a.Definition.identityResolver = nil + b.Definition.identityResolver = nil + + defer func() { + a.Definition.operationUnmarshaler = backOpUnA + b.Definition.operationUnmarshaler = backOpUnB + a.Definition.identityResolver = backIdResA + b.Definition.identityResolver = backIdResB + }() + + require.Equal(t, a, b) +} + +// // Merge +// +// merge1 := makeCommit(t, repo) +// merge1 = makeCommit(t, repo, merge1) +// err = repo.UpdateRef("merge1", merge1) +// require.NoError(t, err) +// +// err = repo.UpdateRef("merge2", merge1) +// require.NoError(t, err) +// +// // identical merge +// err = repo.MergeRef("merge1", "merge2") +// require.NoError(t, err) +// +// refMerge1, err := repo.ResolveRef("merge1") +// require.NoError(t, err) +// require.Equal(t, merge1, refMerge1) +// refMerge2, err := repo.ResolveRef("merge2") +// require.NoError(t, err) +// require.Equal(t, merge1, refMerge2) +// +// // fast-forward merge +// merge2 := makeCommit(t, repo, merge1) +// merge2 = makeCommit(t, repo, merge2) +// +// err = repo.UpdateRef("merge2", merge2) +// require.NoError(t, err) +// +// err = repo.MergeRef("merge1", "merge2") +// require.NoError(t, err) +// +// refMerge1, err = repo.ResolveRef("merge1") +// require.NoError(t, err) +// require.Equal(t, merge2, refMerge1) +// refMerge2, err = repo.ResolveRef("merge2") +// require.NoError(t, err) +// require.Equal(t, merge2, refMerge2) +// +// // merge commit +// merge1 = makeCommit(t, repo, merge1) +// err = repo.UpdateRef("merge1", merge1) +// require.NoError(t, err) +// +// merge2 = makeCommit(t, repo, merge2) +// err = repo.UpdateRef("merge2", merge2) +// require.NoError(t, err) +// +// err = repo.MergeRef("merge1", "merge2") +// require.NoError(t, err) +// +// refMerge1, err = repo.ResolveRef("merge1") +// require.NoError(t, err) +// require.NotEqual(t, merge1, refMerge1) +// commitRefMerge1, err := repo.ReadCommit(refMerge1) +// require.NoError(t, err) +// require.ElementsMatch(t, commitRefMerge1.Parents, []Hash{merge1, merge2}) +// refMerge2, err = repo.ResolveRef("merge2") +// require.NoError(t, err) +// require.Equal(t, merge2, refMerge2) diff --git a/entity/dag/operation.go b/entity/dag/operation.go new file mode 100644 index 00000000..9fcc055b --- /dev/null +++ b/entity/dag/operation.go @@ -0,0 +1,31 @@ +package dag + +import ( + "github.com/MichaelMure/git-bug/entity" + "github.com/MichaelMure/git-bug/identity" +) + +// Operation is a piece of data defining a change to reflect on the state of an Entity. +// What this Operation or Entity's state looks like is not of the resort of this package as it only deals with the +// data structure and storage. +type Operation interface { + // Id return the Operation identifier + // Some care need to be taken to define a correct Id derivation and enough entropy in the data used to avoid + // collisions. Notably: + // - the Id of the first Operation will be used as the Id of the Entity. Collision need to be avoided across Entities. + // - collisions can also happen within the set of Operations of an Entity. Simple Operation might not have enough + // entropy to yield unique Ids. + // A common way to derive an Id will be to use the DeriveId function on the serialized operation data. + Id() entity.Id + // Validate check if the Operation data is valid + Validate() error + + Author() identity.Interface +} + +type operationBase struct { + author identity.Interface + + // Not serialized. Store the op's id in memory. + id entity.Id +} diff --git a/entity/dag/operation_pack.go b/entity/dag/operation_pack.go new file mode 100644 index 00000000..7cf4ee58 --- /dev/null +++ b/entity/dag/operation_pack.go @@ -0,0 +1,294 @@ +package dag + +import ( + "encoding/json" + "fmt" + "strconv" + "strings" + + "github.com/pkg/errors" + "golang.org/x/crypto/openpgp" + + "github.com/MichaelMure/git-bug/entity" + "github.com/MichaelMure/git-bug/identity" + "github.com/MichaelMure/git-bug/repository" + "github.com/MichaelMure/git-bug/util/lamport" +) + +// TODO: extra data tree +const extraEntryName = "extra" + +const opsEntryName = "ops" +const versionEntryPrefix = "version-" +const createClockEntryPrefix = "create-clock-" +const editClockEntryPrefix = "edit-clock-" +const packClockEntryPrefix = "pack-clock-" + +// operationPack is a wrapper structure to store multiple operations in a single git blob. +// Additionally, it holds and store the metadata for those operations. +type operationPack struct { + // An identifier, taken from a hash of the serialized Operations. + id entity.Id + + // The author of the Operations. Must be the same author for all the Operations. + Author identity.Interface + // The list of Operation stored in the operationPack + Operations []Operation + // Encode the entity's logical time of creation across all entities of the same type. + // Only exist on the root operationPack + CreateTime lamport.Time + // Encode the entity's logical time of last edition across all entities of the same type. + // Exist on all operationPack + EditTime lamport.Time + // // Encode the operationPack's logical time of creation withing this entity. + // // Exist on all operationPack + // PackTime lamport.Time +} + +func (opp *operationPack) Id() entity.Id { + if opp.id == "" || opp.id == entity.UnsetId { + // This means we are trying to get the opp's Id *before* it has been stored. + // As the Id is computed based on the actual bytes written on the disk, we are going to predict + // those and then get the Id. This is safe as it will be the exact same code writing on disk later. + + data, err := json.Marshal(opp) + if err != nil { + panic(err) + } + opp.id = entity.DeriveId(data) + } + + return opp.id +} + +func (opp *operationPack) MarshalJSON() ([]byte, error) { + return json.Marshal(struct { + Author identity.Interface `json:"author"` + Operations []Operation `json:"ops"` + }{ + Author: opp.Author, + Operations: opp.Operations, + }) +} + +func (opp *operationPack) Validate() error { + if opp.Author == nil { + return fmt.Errorf("missing author") + } + for _, op := range opp.Operations { + if op.Author() != opp.Author { + return fmt.Errorf("operation has different author than the operationPack's") + } + } + if opp.EditTime == 0 { + return fmt.Errorf("lamport edit time is zero") + } + return nil +} + +func (opp *operationPack) Write(def Definition, repo repository.RepoData, parentCommit ...repository.Hash) (repository.Hash, error) { + if err := opp.Validate(); err != nil { + return "", err + } + + // For different reason, we store the clocks and format version directly in the git tree. + // Version has to be accessible before any attempt to decode to return early with a unique error. + // Clocks could possibly be stored in the git blob but it's nice to separate data and metadata, and + // we are storing something directly in the tree already so why not. + // + // To have a valid Tree, we point the "fake" entries to always the same value, the empty blob. + emptyBlobHash, err := repo.StoreData([]byte{}) + if err != nil { + return "", err + } + + // Write the Ops as a Git blob containing the serialized array of operations + data, err := json.Marshal(opp) + if err != nil { + return "", err + } + + // compute the Id while we have the serialized data + opp.id = entity.DeriveId(data) + + hash, err := repo.StoreData(data) + if err != nil { + return "", err + } + + // Make a Git tree referencing this blob and encoding the other values: + // - format version + // - clocks + tree := []repository.TreeEntry{ + {ObjectType: repository.Blob, Hash: emptyBlobHash, + Name: fmt.Sprintf(versionEntryPrefix+"%d", def.formatVersion)}, + {ObjectType: repository.Blob, Hash: hash, + Name: opsEntryName}, + {ObjectType: repository.Blob, Hash: emptyBlobHash, + Name: fmt.Sprintf(editClockEntryPrefix+"%d", opp.EditTime)}, + // {ObjectType: repository.Blob, Hash: emptyBlobHash, + // Name: fmt.Sprintf(packClockEntryPrefix+"%d", opp.PackTime)}, + } + if opp.CreateTime > 0 { + tree = append(tree, repository.TreeEntry{ + ObjectType: repository.Blob, + Hash: emptyBlobHash, + Name: fmt.Sprintf(createClockEntryPrefix+"%d", opp.CreateTime), + }) + } + + // Store the tree + treeHash, err := repo.StoreTree(tree) + if err != nil { + return "", err + } + + // Write a Git commit referencing the tree, with the previous commit as parent + // If we have keys, sign. + var commitHash repository.Hash + + // Sign the commit if we have a key + if opp.Author.SigningKey() != nil { + commitHash, err = repo.StoreSignedCommit(treeHash, opp.Author.SigningKey().PGPEntity(), parentCommit...) + } else { + commitHash, err = repo.StoreCommit(treeHash, parentCommit...) + } + + if err != nil { + return "", err + } + + return commitHash, nil +} + +// readOperationPack read the operationPack encoded in git at the given Tree hash. +// +// Validity of the Lamport clocks is left for the caller to decide. +func readOperationPack(def Definition, repo repository.RepoData, commit repository.Commit) (*operationPack, error) { + entries, err := repo.ReadTree(commit.TreeHash) + if err != nil { + return nil, err + } + + // check the format version first, fail early instead of trying to read something + var version uint + for _, entry := range entries { + if strings.HasPrefix(entry.Name, versionEntryPrefix) { + v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, versionEntryPrefix), 10, 64) + if err != nil { + return nil, errors.Wrap(err, "can't read format version") + } + if v > 1<<12 { + return nil, fmt.Errorf("format version too big") + } + version = uint(v) + break + } + } + if version == 0 { + return nil, entity.NewErrUnknowFormat(def.formatVersion) + } + if version != def.formatVersion { + return nil, entity.NewErrInvalidFormat(version, def.formatVersion) + } + + var id entity.Id + var author identity.Interface + var ops []Operation + var createTime lamport.Time + var editTime lamport.Time + // var packTime lamport.Time + + for _, entry := range entries { + switch { + case entry.Name == opsEntryName: + data, err := repo.ReadData(entry.Hash) + if err != nil { + return nil, errors.Wrap(err, "failed to read git blob data") + } + ops, author, err = unmarshallPack(def, data) + if err != nil { + return nil, err + } + id = entity.DeriveId(data) + + case strings.HasPrefix(entry.Name, createClockEntryPrefix): + v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, createClockEntryPrefix), 10, 64) + if err != nil { + return nil, errors.Wrap(err, "can't read creation lamport time") + } + createTime = lamport.Time(v) + + case strings.HasPrefix(entry.Name, editClockEntryPrefix): + v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, editClockEntryPrefix), 10, 64) + if err != nil { + return nil, errors.Wrap(err, "can't read edit lamport time") + } + editTime = lamport.Time(v) + + // case strings.HasPrefix(entry.Name, packClockEntryPrefix): + // found &= 1 << 3 + // + // v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, packClockEntryPrefix), 10, 64) + // if err != nil { + // return nil, errors.Wrap(err, "can't read pack lamport time") + // } + // packTime = lamport.Time(v) + } + } + + // Verify signature if we expect one + keys := author.ValidKeysAtTime(fmt.Sprintf(editClockPattern, def.namespace), editTime) + if len(keys) > 0 { + keyring := identity.PGPKeyring(keys) + _, err = openpgp.CheckDetachedSignature(keyring, commit.SignedData, commit.Signature) + if err != nil { + return nil, fmt.Errorf("signature failure: %v", err) + } + } + + return &operationPack{ + id: id, + Author: author, + Operations: ops, + CreateTime: createTime, + EditTime: editTime, + // PackTime: packTime, + }, nil +} + +// unmarshallPack delegate the unmarshalling of the Operation's JSON to the decoding +// function provided by the concrete entity. This gives access to the concrete type of each +// Operation. +func unmarshallPack(def Definition, data []byte) ([]Operation, identity.Interface, error) { + aux := struct { + Author identity.IdentityStub `json:"author"` + Operations []json.RawMessage `json:"ops"` + }{} + + if err := json.Unmarshal(data, &aux); err != nil { + return nil, nil, err + } + + if aux.Author.Id() == "" || aux.Author.Id() == entity.UnsetId { + return nil, nil, fmt.Errorf("missing author") + } + + author, err := def.identityResolver.ResolveIdentity(aux.Author.Id()) + if err != nil { + return nil, nil, err + } + + ops := make([]Operation, 0, len(aux.Operations)) + + for _, raw := range aux.Operations { + // delegate to specialized unmarshal function + op, err := def.operationUnmarshaler(author, raw) + if err != nil { + return nil, nil, err + } + ops = append(ops, op) + } + + return ops, author, nil +} diff --git a/entity/dag/operation_pack_test.go b/entity/dag/operation_pack_test.go new file mode 100644 index 00000000..ad2a9859 --- /dev/null +++ b/entity/dag/operation_pack_test.go @@ -0,0 +1,44 @@ +package dag + +import ( + "testing" + + "github.com/stretchr/testify/require" +) + +func TestOperationPackReadWrite(t *testing.T) { + repo, id1, _, def := makeTestContext() + + opp := &operationPack{ + Author: id1, + Operations: []Operation{ + newOp1(id1, "foo"), + newOp2(id1, "bar"), + }, + CreateTime: 123, + EditTime: 456, + } + + commitHash, err := opp.Write(def, repo) + require.NoError(t, err) + + commit, err := repo.ReadCommit(commitHash) + require.NoError(t, err) + + opp2, err := readOperationPack(def, repo, commit) + require.NoError(t, err) + + require.Equal(t, opp, opp2) + + // make sure we get the same Id with the same data + opp3 := &operationPack{ + Author: id1, + Operations: []Operation{ + newOp1(id1, "foo"), + newOp2(id1, "bar"), + }, + CreateTime: 123, + EditTime: 456, + } + require.Equal(t, opp.Id(), opp3.Id()) +} diff --git a/entity/doc.go b/entity/doc.go deleted file mode 100644 index 4682d545..00000000 --- a/entity/doc.go +++ /dev/null @@ -1,8 +0,0 @@ -// Package entity contains the base common code to define an entity stored -// in a chain of git objects, supporting actions like Push, Pull and Merge. -package entity - -// TODO: Bug and Identity are very similar, right ? I expect that this package -// will eventually hold the common code to define an entity and the related -// helpers, errors and so on. When this work is done, it will become easier -// to add new entities, for example to support pull requests. diff --git a/entity/entity_actions.go b/entity/entity_actions.go deleted file mode 100644 index 34e76a62..00000000 --- a/entity/entity_actions.go +++ /dev/null @@ -1,31 +0,0 @@ -package entity - -import ( - "fmt" - - "github.com/MichaelMure/git-bug/repository" -) - -func ListLocalIds(typename string, repo repository.RepoData) ([]Id, error) { - refs, err := repo.ListRefs(fmt.Sprintf("refs/%s/", typename)) - if err != nil { - return nil, err - } - return RefsToIds(refs), nil -} - -func Fetch() { - -} - -func Pull() { - -} - -func Push() { - -} - -func Remove() error { - panic("") -} diff --git a/entity/entity_test.go b/entity/entity_test.go deleted file mode 100644 index 92a53179..00000000 --- a/entity/entity_test.go +++ /dev/null @@ -1,107 +0,0 @@ -package entity - -import ( - "encoding/json" - "fmt" - "testing" - - "github.com/stretchr/testify/require" - - "github.com/MichaelMure/git-bug/repository" -) - -// func TestFoo(t *testing.T) { -// repo, err := repository.OpenGoGitRepo("~/dev/git-bug", nil) -// require.NoError(t, err) -// -// b, err := ReadBug(repo, Id("8b22e548c93a6ed23c31fd4e337c6286c3d1e5c9cae5537bc8e5842e11bd1099")) -// require.NoError(t, err) -// -// fmt.Println(b) -// } - -type op1 struct { - OperationType int `json:"type"` - Field1 string `json:"field_1"` -} - -func newOp1(field1 string) *op1 { - return &op1{OperationType: 1, Field1: field1} -} - -func (o op1) Id() Id { - data, _ := json.Marshal(o) - return DeriveId(data) -} - -func (o op1) Validate() error { return nil } - -type op2 struct { - OperationType int `json:"type"` - Field2 string `json:"field_2"` -} - -func newOp2(field2 string) *op2 { - return &op2{OperationType: 2, Field2: field2} -} - -func (o op2) Id() Id { - data, _ := json.Marshal(o) - return DeriveId(data) -} - -func (o op2) Validate() error { return nil } - -var def = Definition{ - typename: "foo", - namespace: "foos", - operationUnmarshaler: unmarshaller, - formatVersion: 1, -} - -func unmarshaller(raw json.RawMessage) (Operation, error) { - var t struct { - OperationType int `json:"type"` - } - - if err := json.Unmarshal(raw, &t); err != nil { - return nil, err - } - - switch t.OperationType { - case 1: - op := &op1{} - err := json.Unmarshal(raw, &op) - return op, err - case 2: - op := &op2{} - err := json.Unmarshal(raw, &op) - return op, err - default: - return nil, fmt.Errorf("unknown operation type %v", t.OperationType) - } -} - -func TestWriteRead(t *testing.T) { - repo := repository.NewMockRepo() - - entity := New(def) - require.False(t, entity.NeedCommit()) - - entity.Append(newOp1("foo")) - entity.Append(newOp2("bar")) - - require.True(t, entity.NeedCommit()) - require.NoError(t, entity.CommitAdNeeded(repo)) - require.False(t, entity.NeedCommit()) - - entity.Append(newOp2("foobar")) - require.True(t, entity.NeedCommit()) - require.NoError(t, entity.CommitAdNeeded(repo)) - require.False(t, entity.NeedCommit()) - - read, err := Read(def, repo, entity.Id()) - require.NoError(t, err) - - fmt.Println(*read) -} diff --git a/entity/merge.go b/entity/merge.go index 3ce8edac..7d1f3f43 100644 --- a/entity/merge.go +++ b/entity/merge.go @@ -8,14 +8,15 @@ import ( type MergeStatus int const ( - _ MergeStatus = iota - MergeStatusNew - MergeStatusInvalid - MergeStatusUpdated - MergeStatusNothing - MergeStatusError + _ MergeStatus = iota + MergeStatusNew // a new Entity was created locally + MergeStatusInvalid // the remote data is invalid + MergeStatusUpdated // a local Entity has been updated + MergeStatusNothing // no changes were made to a local Entity (already up to date) + MergeStatusError // a terminal error happened ) +// MergeResult hold the result of a merge operation on an Entity. type MergeResult struct { // Err is set when a terminal error occur in the process Err error @@ -55,6 +56,7 @@ func NewMergeError(err error, id Id) MergeResult { } } +// TODO: Interface --> *Entity ? func NewMergeStatus(status MergeStatus, id Id, entity Interface) MergeResult { return MergeResult{ Id: id, diff --git a/entity/operation_pack.go b/entity/operation_pack.go deleted file mode 100644 index 0a16dd61..00000000 --- a/entity/operation_pack.go +++ /dev/null @@ -1,199 +0,0 @@ -package entity - -import ( - "encoding/json" - "fmt" - "strconv" - "strings" - - "github.com/pkg/errors" - - "github.com/MichaelMure/git-bug/repository" - "github.com/MichaelMure/git-bug/util/lamport" -) - -// TODO: extra data tree -const extraEntryName = "extra" - -const opsEntryName = "ops" -const versionEntryPrefix = "version-" -const createClockEntryPrefix = "create-clock-" -const editClockEntryPrefix = "edit-clock-" -const packClockEntryPrefix = "pack-clock-" - -type operationPack struct { - Operations []Operation - // Encode the entity's logical time of creation across all entities of the same type. - // Only exist on the root operationPack - CreateTime lamport.Time - // Encode the entity's logical time of last edition across all entities of the same type. - // Exist on all operationPack - EditTime lamport.Time - // Encode the operationPack's logical time of creation withing this entity. - // Exist on all operationPack - PackTime lamport.Time -} - -func (opp operationPack) write(def Definition, repo repository.RepoData) (repository.Hash, error) { - // For different reason, we store the clocks and format version directly in the git tree. - // Version has to be accessible before any attempt to decode to return early with a unique error. - // Clocks could possibly be stored in the git blob but it's nice to separate data and metadata, and - // we are storing something directly in the tree already so why not. - // - // To have a valid Tree, we point the "fake" entries to always the same value, the empty blob. - emptyBlobHash, err := repo.StoreData([]byte{}) - if err != nil { - return "", err - } - - // Write the Ops as a Git blob containing the serialized array - data, err := json.Marshal(struct { - Operations []Operation `json:"ops"` - }{ - Operations: opp.Operations, - }) - if err != nil { - return "", err - } - hash, err := repo.StoreData(data) - if err != nil { - return "", err - } - - // Make a Git tree referencing this blob and encoding the other values: - // - format version - // - clocks - tree := []repository.TreeEntry{ - {ObjectType: repository.Blob, Hash: emptyBlobHash, - Name: fmt.Sprintf(versionEntryPrefix+"%d", def.formatVersion)}, - {ObjectType: repository.Blob, Hash: hash, - Name: opsEntryName}, - {ObjectType: repository.Blob, Hash: emptyBlobHash, - Name: fmt.Sprintf(editClockEntryPrefix+"%d", opp.EditTime)}, - {ObjectType: repository.Blob, Hash: emptyBlobHash, - Name: fmt.Sprintf(packClockEntryPrefix+"%d", opp.PackTime)}, - } - if opp.CreateTime > 0 { - tree = append(tree, repository.TreeEntry{ - ObjectType: repository.Blob, - Hash: emptyBlobHash, - Name: fmt.Sprintf(createClockEntryPrefix+"%d", opp.CreateTime), - }) - } - - // Store the tree - return repo.StoreTree(tree) -} - -// readOperationPack read the operationPack encoded in git at the given Tree hash. -// -// Validity of the Lamport clocks is left for the caller to decide. -func readOperationPack(def Definition, repo repository.RepoData, treeHash repository.Hash) (*operationPack, error) { - entries, err := repo.ReadTree(treeHash) - if err != nil { - return nil, err - } - - // check the format version first, fail early instead of trying to read something - var version uint - for _, entry := range entries { - if strings.HasPrefix(entry.Name, versionEntryPrefix) { - v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, versionEntryPrefix), 10, 64) - if err != nil { - return nil, errors.Wrap(err, "can't read format version") - } - if v > 1<<12 { - return nil, fmt.Errorf("format version too big") - } - version = uint(v) - break - } - } - if version == 0 { - return nil, NewErrUnknowFormat(def.formatVersion) - } - if version != def.formatVersion { - return nil, NewErrInvalidFormat(version, def.formatVersion) - } - - var ops []Operation - var createTime lamport.Time - var editTime lamport.Time - var packTime lamport.Time - - for _, entry := range entries { - if entry.Name == opsEntryName { - data, err := repo.ReadData(entry.Hash) - if err != nil { - return nil, errors.Wrap(err, "failed to read git blob data") - } - - ops, err = unmarshallOperations(def, data) - if err != nil { - return nil, err - } - continue - } - - if strings.HasPrefix(entry.Name, createClockEntryPrefix) { - v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, createClockEntryPrefix), 10, 64) - if err != nil { - return nil, errors.Wrap(err, "can't read creation lamport time") - } - createTime = lamport.Time(v) - continue - } - - if strings.HasPrefix(entry.Name, editClockEntryPrefix) { - v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, editClockEntryPrefix), 10, 64) - if err != nil { - return nil, errors.Wrap(err, "can't read edit lamport time") - } - editTime = lamport.Time(v) - continue - } - - if strings.HasPrefix(entry.Name, packClockEntryPrefix) { - v, err := strconv.ParseUint(strings.TrimPrefix(entry.Name, packClockEntryPrefix), 10, 64) - if err != nil { - return nil, errors.Wrap(err, "can't read pack lamport time") - } - packTime = lamport.Time(v) - continue - } - } - - return &operationPack{ - Operations: ops, - CreateTime: createTime, - EditTime: editTime, - PackTime: packTime, - }, nil -} - -// unmarshallOperations delegate the unmarshalling of the Operation's JSON to the decoding -// function provided by the concrete entity. This gives access to the concrete type of each -// Operation. -func unmarshallOperations(def Definition, data []byte) ([]Operation, error) { - aux := struct { - Operations []json.RawMessage `json:"ops"` - }{} - - if err := json.Unmarshal(data, &aux); err != nil { - return nil, err - } - - ops := make([]Operation, 0, len(aux.Operations)) - - for _, raw := range aux.Operations { - // delegate to specialized unmarshal function - op, err := def.operationUnmarshaler(raw) - if err != nil { - return nil, err - } - - ops = append(ops, op) - } - - return ops, nil -} diff --git a/entity/refs.go b/entity/refs.go index f505dbf0..070d4dba 100644 --- a/entity/refs.go +++ b/entity/refs.go @@ -2,6 +2,7 @@ package entity import "strings" +// RefsToIds parse a slice of git references and return the corresponding Entity's Id. func RefsToIds(refs []string) []Id { ids := make([]Id, len(refs)) @@ -12,6 +13,7 @@ func RefsToIds(refs []string) []Id { return ids } +// RefsToIds parse a git reference and return the corresponding Entity's Id. func RefToId(ref string) Id { split := strings.Split(ref, "/") return Id(split[len(split)-1]) diff --git a/identity/identity.go b/identity/identity.go index ef488712..65019041 100644 --- a/identity/identity.go +++ b/identity/identity.go @@ -344,7 +344,7 @@ func (i *Identity) Commit(repo repository.ClockedRepo) error { var commitHash repository.Hash if lastCommit != "" { - commitHash, err = repo.StoreCommitWithParent(treeHash, lastCommit) + commitHash, err = repo.StoreCommit(treeHash, lastCommit) } else { commitHash, err = repo.StoreCommit(treeHash) } @@ -518,6 +518,15 @@ func (i *Identity) Keys() []*Key { return i.lastVersion().keys } +// SigningKey return the key that should be used to sign new messages. If no key is available, return nil. +func (i *Identity) SigningKey() *Key { + keys := i.Keys() + if len(keys) > 0 { + return keys[0] + } + return nil +} + // ValidKeysAtTime return the set of keys valid at a given lamport time func (i *Identity) ValidKeysAtTime(clockName string, time lamport.Time) []*Key { var result []*Key diff --git a/identity/identity_actions_test.go b/identity/identity_actions_test.go index 63f6aacd..54cb2a46 100644 --- a/identity/identity_actions_test.go +++ b/identity/identity_actions_test.go @@ -9,7 +9,7 @@ import ( ) func TestPushPull(t *testing.T) { - repoA, repoB, remote := repository.SetupReposAndRemote() + repoA, repoB, remote := repository.SetupGoGitReposAndRemote() defer repository.CleanupTestRepos(repoA, repoB, remote) identity1, err := NewIdentity(repoA, "name1", "email1") diff --git a/identity/identity_stub.go b/identity/identity_stub.go index fec92010..91945378 100644 --- a/identity/identity_stub.go +++ b/identity/identity_stub.go @@ -71,6 +71,10 @@ func (IdentityStub) Keys() []*Key { panic("identities needs to be properly loaded with identity.ReadLocal()") } +func (i *IdentityStub) SigningKey() *Key { + panic("identities needs to be properly loaded with identity.ReadLocal()") +} + func (IdentityStub) ValidKeysAtTime(_ string, _ lamport.Time) []*Key { panic("identities needs to be properly loaded with identity.ReadLocal()") } diff --git a/identity/identity_test.go b/identity/identity_test.go index ad8317ce..2cdb4b36 100644 --- a/identity/identity_test.go +++ b/identity/identity_test.go @@ -36,18 +36,18 @@ func TestIdentityCommitLoad(t *testing.T) { // multiple versions - identity, err = NewIdentityFull(repo, "René Descartes", "rene.descartes@example.com", "", "", []*Key{{PubKey: "pubkeyA"}}) + identity, err = NewIdentityFull(repo, "René Descartes", "rene.descartes@example.com", "", "", []*Key{generatePublicKey()}) require.NoError(t, err) idBeforeCommit = identity.Id() err = identity.Mutate(repo, func(orig *Mutator) { - orig.Keys = []*Key{{PubKey: "pubkeyB"}} + orig.Keys = []*Key{generatePublicKey()} }) require.NoError(t, err) err = identity.Mutate(repo, func(orig *Mutator) { - orig.Keys = []*Key{{PubKey: "pubkeyC"}} + orig.Keys = []*Key{generatePublicKey()} }) require.NoError(t, err) @@ -70,13 +70,13 @@ func TestIdentityCommitLoad(t *testing.T) { err = identity.Mutate(repo, func(orig *Mutator) { orig.Email = "rene@descartes.com" - orig.Keys = []*Key{{PubKey: "pubkeyD"}} + orig.Keys = []*Key{generatePublicKey()} }) require.NoError(t, err) err = identity.Mutate(repo, func(orig *Mutator) { orig.Email = "rene@descartes.com" - orig.Keys = []*Key{{PubKey: "pubkeyD"}, {PubKey: "pubkeyE"}} + orig.Keys = []*Key{generatePublicKey(), generatePublicKey()} }) require.NoError(t, err) @@ -123,49 +123,45 @@ func commitsAreSet(t *testing.T, identity *Identity) { // Test that the correct crypto keys are returned for a given lamport time func TestIdentity_ValidKeysAtTime(t *testing.T) { + pubKeyA := generatePublicKey() + pubKeyB := generatePublicKey() + pubKeyC := generatePublicKey() + pubKeyD := generatePublicKey() + pubKeyE := generatePublicKey() + identity := Identity{ versions: []*version{ { times: map[string]lamport.Time{"foo": 100}, - keys: []*Key{ - {PubKey: "pubkeyA"}, - }, + keys: []*Key{pubKeyA}, }, { times: map[string]lamport.Time{"foo": 200}, - keys: []*Key{ - {PubKey: "pubkeyB"}, - }, + keys: []*Key{pubKeyB}, }, { times: map[string]lamport.Time{"foo": 201}, - keys: []*Key{ - {PubKey: "pubkeyC"}, - }, + keys: []*Key{pubKeyC}, }, { times: map[string]lamport.Time{"foo": 201}, - keys: []*Key{ - {PubKey: "pubkeyD"}, - }, + keys: []*Key{pubKeyD}, }, { times: map[string]lamport.Time{"foo": 300}, - keys: []*Key{ - {PubKey: "pubkeyE"}, - }, + keys: []*Key{pubKeyE}, }, }, } require.Nil(t, identity.ValidKeysAtTime("foo", 10)) - require.Equal(t, identity.ValidKeysAtTime("foo", 100), []*Key{{PubKey: "pubkeyA"}}) - require.Equal(t, identity.ValidKeysAtTime("foo", 140), []*Key{{PubKey: "pubkeyA"}}) - require.Equal(t, identity.ValidKeysAtTime("foo", 200), []*Key{{PubKey: "pubkeyB"}}) - require.Equal(t, identity.ValidKeysAtTime("foo", 201), []*Key{{PubKey: "pubkeyD"}}) - require.Equal(t, identity.ValidKeysAtTime("foo", 202), []*Key{{PubKey: "pubkeyD"}}) - require.Equal(t, identity.ValidKeysAtTime("foo", 300), []*Key{{PubKey: "pubkeyE"}}) - require.Equal(t, identity.ValidKeysAtTime("foo", 3000), []*Key{{PubKey: "pubkeyE"}}) + require.Equal(t, identity.ValidKeysAtTime("foo", 100), []*Key{pubKeyA}) + require.Equal(t, identity.ValidKeysAtTime("foo", 140), []*Key{pubKeyA}) + require.Equal(t, identity.ValidKeysAtTime("foo", 200), []*Key{pubKeyB}) + require.Equal(t, identity.ValidKeysAtTime("foo", 201), []*Key{pubKeyD}) + require.Equal(t, identity.ValidKeysAtTime("foo", 202), []*Key{pubKeyD}) + require.Equal(t, identity.ValidKeysAtTime("foo", 300), []*Key{pubKeyE}) + require.Equal(t, identity.ValidKeysAtTime("foo", 3000), []*Key{pubKeyE}) } // Test the immutable or mutable metadata search diff --git a/identity/interface.go b/identity/interface.go index 92a03c51..528cb067 100644 --- a/identity/interface.go +++ b/identity/interface.go @@ -36,6 +36,9 @@ type Interface interface { // Can be empty. Keys() []*Key + // SigningKey return the key that should be used to sign new messages. If no key is available, return nil. + SigningKey() *Key + // ValidKeysAtTime return the set of keys valid at a given lamport time for a given clock of another entity // Can be empty. ValidKeysAtTime(clockName string, time lamport.Time) []*Key diff --git a/identity/key.go b/identity/key.go index cc948394..8dd5e8c1 100644 --- a/identity/key.go +++ b/identity/key.go @@ -1,18 +1,193 @@ package identity +import ( + "bytes" + "encoding/json" + "fmt" + "io" + "strings" + "time" + + "github.com/pkg/errors" + "golang.org/x/crypto/openpgp" + "golang.org/x/crypto/openpgp/armor" + "golang.org/x/crypto/openpgp/packet" + + "github.com/MichaelMure/git-bug/repository" +) + type Key struct { - // The GPG fingerprint of the key - Fingerprint string `json:"fingerprint"` - PubKey string `json:"pub_key"` + public *packet.PublicKey + private *packet.PrivateKey +} + +// GenerateKey generate a keypair (public+private) +func GenerateKey() *Key { + entity, err := openpgp.NewEntity("", "", "", &packet.Config{ + // The armored format doesn't include the creation time, which makes the round-trip data not being fully equal. + // We don't care about the creation time so we can set it to the zero value. + Time: func() time.Time { + return time.Time{} + }, + }) + if err != nil { + panic(err) + } + return &Key{ + public: entity.PrimaryKey, + private: entity.PrivateKey, + } +} + +// generatePublicKey generate only a public key (only useful for testing) +// See GenerateKey for the details. +func generatePublicKey() *Key { + k := GenerateKey() + k.private = nil + return k +} + +func (k *Key) MarshalJSON() ([]byte, error) { + var buf bytes.Buffer + w, err := armor.Encode(&buf, openpgp.PublicKeyType, nil) + if err != nil { + return nil, err + } + err = k.public.Serialize(w) + if err != nil { + return nil, err + } + err = w.Close() + if err != nil { + return nil, err + } + return json.Marshal(buf.String()) +} + +func (k *Key) UnmarshalJSON(data []byte) error { + var armored string + err := json.Unmarshal(data, &armored) + if err != nil { + return err + } + + block, err := armor.Decode(strings.NewReader(armored)) + if err == io.EOF { + return fmt.Errorf("no armored data found") + } + if err != nil { + return err + } + + if block.Type != openpgp.PublicKeyType { + return fmt.Errorf("invalid key type") + } + + reader := packet.NewReader(block.Body) + p, err := reader.Next() + if err != nil { + return errors.Wrap(err, "failed to read public key packet") + } + + public, ok := p.(*packet.PublicKey) + if !ok { + return errors.New("got no packet.publicKey") + } + + // The armored format doesn't include the creation time, which makes the round-trip data not being fully equal. + // We don't care about the creation time so we can set it to the zero value. + public.CreationTime = time.Time{} + + k.public = public + return nil } func (k *Key) Validate() error { - // Todo + if k.public == nil { + return fmt.Errorf("nil public key") + } + if !k.public.CanSign() { + return fmt.Errorf("public key can't sign") + } + + if k.private != nil { + if !k.private.CanSign() { + return fmt.Errorf("private key can't sign") + } + } return nil } func (k *Key) Clone() *Key { - clone := *k - return &clone + clone := &Key{} + + pub := *k.public + clone.public = &pub + + if k.private != nil { + priv := *k.private + clone.private = &priv + } + + return clone +} + +func (k *Key) EnsurePrivateKey(repo repository.RepoKeyring) error { + if k.private != nil { + return nil + } + + // item, err := repo.Keyring().Get(k.Fingerprint()) + // if err != nil { + // return fmt.Errorf("no private key found for %s", k.Fingerprint()) + // } + // + + panic("TODO") +} + +func (k *Key) Fingerprint() string { + return string(k.public.Fingerprint[:]) +} + +func (k *Key) PGPEntity() *openpgp.Entity { + return &openpgp.Entity{ + PrimaryKey: k.public, + PrivateKey: k.private, + } +} + +var _ openpgp.KeyRing = &PGPKeyring{} + +// PGPKeyring implement a openpgp.KeyRing from an slice of Key +type PGPKeyring []*Key + +func (pk PGPKeyring) KeysById(id uint64) []openpgp.Key { + var result []openpgp.Key + for _, key := range pk { + if key.public.KeyId == id { + result = append(result, openpgp.Key{ + PublicKey: key.public, + PrivateKey: key.private, + }) + } + } + return result +} + +func (pk PGPKeyring) KeysByIdUsage(id uint64, requiredUsage byte) []openpgp.Key { + // the only usage we care about is the ability to sign, which all keys should already be capable of + return pk.KeysById(id) +} + +func (pk PGPKeyring) DecryptionKeys() []openpgp.Key { + result := make([]openpgp.Key, len(pk)) + for i, key := range pk { + result[i] = openpgp.Key{ + PublicKey: key.public, + PrivateKey: key.private, + } + } + return result } diff --git a/identity/key_test.go b/identity/key_test.go new file mode 100644 index 00000000..3206c34e --- /dev/null +++ b/identity/key_test.go @@ -0,0 +1,21 @@ +package identity + +import ( + "encoding/json" + "testing" + + "github.com/stretchr/testify/require" +) + +func TestKeyJSON(t *testing.T) { + k := generatePublicKey() + + data, err := json.Marshal(k) + require.NoError(t, err) + + var read Key + err = json.Unmarshal(data, &read) + require.NoError(t, err) + + require.Equal(t, k, &read) +} diff --git a/identity/version_test.go b/identity/version_test.go index 1efa0d03..385ad4d7 100644 --- a/identity/version_test.go +++ b/identity/version_test.go @@ -18,29 +18,23 @@ func makeIdentityTestRepo(t *testing.T) repository.ClockedRepo { clock1, err := repo.GetOrCreateClock("foo") require.NoError(t, err) - err = clock1.Witness(42) // clock goes to 43 + err = clock1.Witness(42) require.NoError(t, err) clock2, err := repo.GetOrCreateClock("bar") require.NoError(t, err) - err = clock2.Witness(34) // clock goes to 35 + err = clock2.Witness(34) require.NoError(t, err) return repo } -func TestVersionSerialize(t *testing.T) { +func TestVersionJSON(t *testing.T) { repo := makeIdentityTestRepo(t) keys := []*Key{ - { - Fingerprint: "fingerprint1", - PubKey: "pubkey1", - }, - { - Fingerprint: "fingerprint2", - PubKey: "pubkey2", - }, + generatePublicKey(), + generatePublicKey(), } before, err := newVersion(repo, "name", "email", "login", "avatarUrl", keys) @@ -57,8 +51,8 @@ func TestVersionSerialize(t *testing.T) { avatarURL: "avatarUrl", unixTime: time.Now().Unix(), times: map[string]lamport.Time{ - "foo": 43, - "bar": 35, + "foo": 42, + "bar": 34, }, keys: keys, nonce: before.nonce, diff --git a/repository/common.go b/repository/common.go new file mode 100644 index 00000000..7fd7ae19 --- /dev/null +++ b/repository/common.go @@ -0,0 +1,120 @@ +package repository + +import ( + "io" + + "golang.org/x/crypto/openpgp" + "golang.org/x/crypto/openpgp/armor" + "golang.org/x/crypto/openpgp/errors" +) + +// nonNativeMerge is an implementation of a branch merge, for the case where +// the underlying git implementation doesn't support it natively. +func nonNativeMerge(repo RepoData, ref string, otherRef string, treeHashFn func() Hash) error { + commit, err := repo.ResolveRef(ref) + if err != nil { + return err + } + + otherCommit, err := repo.ResolveRef(otherRef) + if err != nil { + return err + } + + if commit == otherCommit { + // nothing to merge + return nil + } + + // fast-forward is possible if otherRef include ref + + otherCommits, err := repo.ListCommits(otherRef) + if err != nil { + return err + } + + fastForwardPossible := false + for _, hash := range otherCommits { + if hash == commit { + fastForwardPossible = true + break + } + } + + if fastForwardPossible { + return repo.UpdateRef(ref, otherCommit) + } + + // fast-forward is not possible, we need to create a merge commit + + // we need a Tree to make the commit, an empty Tree will do + emptyTreeHash, err := repo.StoreTree(nil) + if err != nil { + return err + } + + newHash, err := repo.StoreCommit(emptyTreeHash, commit, otherCommit) + if err != nil { + return err + } + + return repo.UpdateRef(ref, newHash) +} + +// nonNativeListCommits is an implementation for ListCommits, for the case where +// the underlying git implementation doesn't support if natively. +func nonNativeListCommits(repo RepoData, ref string) ([]Hash, error) { + var result []Hash + + stack := make([]Hash, 0, 32) + visited := make(map[Hash]struct{}) + + hash, err := repo.ResolveRef(ref) + if err != nil { + return nil, err + } + + stack = append(stack, hash) + + for len(stack) > 0 { + // pop + hash := stack[len(stack)-1] + stack = stack[:len(stack)-1] + + if _, ok := visited[hash]; ok { + continue + } + + // mark as visited + visited[hash] = struct{}{} + result = append(result, hash) + + commit, err := repo.ReadCommit(hash) + if err != nil { + return nil, err + } + + for _, parent := range commit.Parents { + stack = append(stack, parent) + } + } + + // reverse + for i, j := 0, len(result)-1; i < j; i, j = i+1, j-1 { + result[i], result[j] = result[j], result[i] + } + + return result, nil +} + +// deArmorSignature convert an armored (text serialized) signature into raw binary +func deArmorSignature(armoredSig io.Reader) (io.Reader, error) { + block, err := armor.Decode(armoredSig) + if err != nil { + return nil, err + } + if block.Type != openpgp.SignatureType { + return nil, errors.InvalidArgumentError("expected '" + openpgp.SignatureType + "', got: " + block.Type) + } + return block.Body, nil +} diff --git a/repository/git.go b/repository/git.go deleted file mode 100644 index e89bae87..00000000 --- a/repository/git.go +++ /dev/null @@ -1,570 +0,0 @@ -// Package repository contains helper methods for working with the Git repo. -package repository - -import ( - "bytes" - "fmt" - "io/ioutil" - "os" - "path/filepath" - "strings" - "sync" - - "github.com/blevesearch/bleve" - "github.com/go-git/go-billy/v5" - "github.com/go-git/go-billy/v5/osfs" - - "github.com/MichaelMure/git-bug/util/lamport" -) - -var _ ClockedRepo = &GitRepo{} -var _ TestedRepo = &GitRepo{} - -// GitRepo represents an instance of a (local) git repository. -type GitRepo struct { - gitCli - path string - - clocksMutex sync.Mutex - clocks map[string]lamport.Clock - - indexesMutex sync.Mutex - indexes map[string]bleve.Index - - keyring Keyring - localStorage billy.Filesystem -} - -func (repo *GitRepo) ReadCommit(hash Hash) (Commit, error) { - panic("implement me") -} - -func (repo *GitRepo) ResolveRef(ref string) (Hash, error) { - panic("implement me") -} - -// OpenGitRepo determines if the given working directory is inside of a git repository, -// and returns the corresponding GitRepo instance if it is. -func OpenGitRepo(path string, clockLoaders []ClockLoader) (*GitRepo, error) { - k, err := defaultKeyring() - if err != nil { - return nil, err - } - - repo := &GitRepo{ - gitCli: gitCli{path: path}, - path: path, - clocks: make(map[string]lamport.Clock), - indexes: make(map[string]bleve.Index), - keyring: k, - } - - // Check the repo and retrieve the root path - stdout, err := repo.runGitCommand("rev-parse", "--absolute-git-dir") - - // Now dir is fetched with "git rev-parse --git-dir". May be it can - // still return nothing in some cases. Then empty stdout check is - // kept. - if err != nil || stdout == "" { - return nil, ErrNotARepo - } - - // Fix the path to be sure we are at the root - repo.path = stdout - repo.gitCli.path = stdout - repo.localStorage = osfs.New(filepath.Join(path, "git-bug")) - - for _, loader := range clockLoaders { - allExist := true - for _, name := range loader.Clocks { - if _, err := repo.getClock(name); err != nil { - allExist = false - } - } - - if !allExist { - err = loader.Witnesser(repo) - if err != nil { - return nil, err - } - } - } - - return repo, nil -} - -// InitGitRepo create a new empty git repo at the given path -func InitGitRepo(path string) (*GitRepo, error) { - k, err := defaultKeyring() - if err != nil { - return nil, err - } - - repo := &GitRepo{ - gitCli: gitCli{path: path}, - path: filepath.Join(path, ".git"), - clocks: make(map[string]lamport.Clock), - indexes: make(map[string]bleve.Index), - keyring: k, - localStorage: osfs.New(filepath.Join(path, ".git", "git-bug")), - } - - _, err = repo.runGitCommand("init", path) - if err != nil { - return nil, err - } - - return repo, nil -} - -// InitBareGitRepo create a new --bare empty git repo at the given path -func InitBareGitRepo(path string) (*GitRepo, error) { - k, err := defaultKeyring() - if err != nil { - return nil, err - } - - repo := &GitRepo{ - gitCli: gitCli{path: path}, - path: path, - clocks: make(map[string]lamport.Clock), - indexes: make(map[string]bleve.Index), - keyring: k, - localStorage: osfs.New(filepath.Join(path, "git-bug")), - } - - _, err = repo.runGitCommand("init", "--bare", path) - if err != nil { - return nil, err - } - - return repo, nil -} - -func (repo *GitRepo) Close() error { - var firstErr error - for _, index := range repo.indexes { - err := index.Close() - if err != nil && firstErr == nil { - firstErr = err - } - } - return firstErr -} - -// LocalConfig give access to the repository scoped configuration -func (repo *GitRepo) LocalConfig() Config { - return newGitConfig(repo.gitCli, false) -} - -// GlobalConfig give access to the global scoped configuration -func (repo *GitRepo) GlobalConfig() Config { - return newGitConfig(repo.gitCli, true) -} - -// AnyConfig give access to a merged local/global configuration -func (repo *GitRepo) AnyConfig() ConfigRead { - return mergeConfig(repo.LocalConfig(), repo.GlobalConfig()) -} - -// Keyring give access to a user-wide storage for secrets -func (repo *GitRepo) Keyring() Keyring { - return repo.keyring -} - -// GetPath returns the path to the repo. -func (repo *GitRepo) GetPath() string { - return repo.path -} - -// GetUserName returns the name the the user has used to configure git -func (repo *GitRepo) GetUserName() (string, error) { - return repo.runGitCommand("config", "user.name") -} - -// GetUserEmail returns the email address that the user has used to configure git. -func (repo *GitRepo) GetUserEmail() (string, error) { - return repo.runGitCommand("config", "user.email") -} - -// GetCoreEditor returns the name of the editor that the user has used to configure git. -func (repo *GitRepo) GetCoreEditor() (string, error) { - return repo.runGitCommand("var", "GIT_EDITOR") -} - -// GetRemotes returns the configured remotes repositories. -func (repo *GitRepo) GetRemotes() (map[string]string, error) { - stdout, err := repo.runGitCommand("remote", "--verbose") - if err != nil { - return nil, err - } - - lines := strings.Split(stdout, "\n") - remotes := make(map[string]string, len(lines)) - - for _, line := range lines { - if strings.TrimSpace(line) == "" { - continue - } - elements := strings.Fields(line) - if len(elements) != 3 { - return nil, fmt.Errorf("git remote: unexpected output format: %s", line) - } - - remotes[elements[0]] = elements[1] - } - - return remotes, nil -} - -// LocalStorage return a billy.Filesystem giving access to $RepoPath/.git/git-bug -func (repo *GitRepo) LocalStorage() billy.Filesystem { - return repo.localStorage -} - -// GetBleveIndex return a bleve.Index that can be used to index documents -func (repo *GitRepo) GetBleveIndex(name string) (bleve.Index, error) { - repo.indexesMutex.Lock() - defer repo.indexesMutex.Unlock() - - if index, ok := repo.indexes[name]; ok { - return index, nil - } - - path := filepath.Join(repo.path, "indexes", name) - - index, err := bleve.Open(path) - if err == nil { - repo.indexes[name] = index - return index, nil - } - - err = os.MkdirAll(path, os.ModeDir) - if err != nil { - return nil, err - } - - mapping := bleve.NewIndexMapping() - mapping.DefaultAnalyzer = "en" - - index, err = bleve.New(path, mapping) - if err != nil { - return nil, err - } - - repo.indexes[name] = index - - return index, nil -} - -// ClearBleveIndex will wipe the given index -func (repo *GitRepo) ClearBleveIndex(name string) error { - repo.indexesMutex.Lock() - defer repo.indexesMutex.Unlock() - - path := filepath.Join(repo.path, "indexes", name) - - err := os.RemoveAll(path) - if err != nil { - return err - } - - delete(repo.indexes, name) - - return nil -} - -// FetchRefs fetch git refs from a remote -func (repo *GitRepo) FetchRefs(remote, refSpec string) (string, error) { - stdout, err := repo.runGitCommand("fetch", remote, refSpec) - - if err != nil { - return stdout, fmt.Errorf("failed to fetch from the remote '%s': %v", remote, err) - } - - return stdout, err -} - -// PushRefs push git refs to a remote -func (repo *GitRepo) PushRefs(remote string, refSpec string) (string, error) { - stdout, stderr, err := repo.runGitCommandRaw(nil, "push", remote, refSpec) - - if err != nil { - return stdout + stderr, fmt.Errorf("failed to push to the remote '%s': %v", remote, stderr) - } - return stdout + stderr, nil -} - -// StoreData will store arbitrary data and return the corresponding hash -func (repo *GitRepo) StoreData(data []byte) (Hash, error) { - var stdin = bytes.NewReader(data) - - stdout, err := repo.runGitCommandWithStdin(stdin, "hash-object", "--stdin", "-w") - - return Hash(stdout), err -} - -// ReadData will attempt to read arbitrary data from the given hash -func (repo *GitRepo) ReadData(hash Hash) ([]byte, error) { - var stdout bytes.Buffer - var stderr bytes.Buffer - - err := repo.runGitCommandWithIO(nil, &stdout, &stderr, "cat-file", "-p", string(hash)) - - if err != nil { - return []byte{}, err - } - - return stdout.Bytes(), nil -} - -// StoreTree will store a mapping key-->Hash as a Git tree -func (repo *GitRepo) StoreTree(entries []TreeEntry) (Hash, error) { - buffer := prepareTreeEntries(entries) - - stdout, err := repo.runGitCommandWithStdin(&buffer, "mktree") - - if err != nil { - return "", err - } - - return Hash(stdout), nil -} - -// StoreCommit will store a Git commit with the given Git tree -func (repo *GitRepo) StoreCommit(treeHash Hash) (Hash, error) { - stdout, err := repo.runGitCommand("commit-tree", string(treeHash)) - - if err != nil { - return "", err - } - - return Hash(stdout), nil -} - -// StoreCommitWithParent will store a Git commit with the given Git tree -func (repo *GitRepo) StoreCommitWithParent(treeHash Hash, parent Hash) (Hash, error) { - stdout, err := repo.runGitCommand("commit-tree", string(treeHash), - "-p", string(parent)) - - if err != nil { - return "", err - } - - return Hash(stdout), nil -} - -// UpdateRef will create or update a Git reference -func (repo *GitRepo) UpdateRef(ref string, hash Hash) error { - _, err := repo.runGitCommand("update-ref", ref, string(hash)) - - return err -} - -// RemoveRef will remove a Git reference -func (repo *GitRepo) RemoveRef(ref string) error { - _, err := repo.runGitCommand("update-ref", "-d", ref) - - return err -} - -// ListRefs will return a list of Git ref matching the given refspec -func (repo *GitRepo) ListRefs(refPrefix string) ([]string, error) { - stdout, err := repo.runGitCommand("for-each-ref", "--format=%(refname)", refPrefix) - - if err != nil { - return nil, err - } - - split := strings.Split(stdout, "\n") - - if len(split) == 1 && split[0] == "" { - return []string{}, nil - } - - return split, nil -} - -// RefExist will check if a reference exist in Git -func (repo *GitRepo) RefExist(ref string) (bool, error) { - stdout, err := repo.runGitCommand("for-each-ref", ref) - - if err != nil { - return false, err - } - - return stdout != "", nil -} - -// CopyRef will create a new reference with the same value as another one -func (repo *GitRepo) CopyRef(source string, dest string) error { - _, err := repo.runGitCommand("update-ref", dest, source) - - return err -} - -// ListCommits will return the list of commit hashes of a ref, in chronological order -func (repo *GitRepo) ListCommits(ref string) ([]Hash, error) { - stdout, err := repo.runGitCommand("rev-list", "--first-parent", "--reverse", ref) - - if err != nil { - return nil, err - } - - split := strings.Split(stdout, "\n") - - casted := make([]Hash, len(split)) - for i, line := range split { - casted[i] = Hash(line) - } - - return casted, nil - -} - -// ReadTree will return the list of entries in a Git tree -func (repo *GitRepo) ReadTree(hash Hash) ([]TreeEntry, error) { - stdout, err := repo.runGitCommand("ls-tree", string(hash)) - - if err != nil { - return nil, err - } - - return readTreeEntries(stdout) -} - -// FindCommonAncestor will return the last common ancestor of two chain of commit -func (repo *GitRepo) FindCommonAncestor(hash1 Hash, hash2 Hash) (Hash, error) { - stdout, err := repo.runGitCommand("merge-base", string(hash1), string(hash2)) - - if err != nil { - return "", err - } - - return Hash(stdout), nil -} - -// GetTreeHash return the git tree hash referenced in a commit -func (repo *GitRepo) GetTreeHash(commit Hash) (Hash, error) { - stdout, err := repo.runGitCommand("rev-parse", string(commit)+"^{tree}") - - if err != nil { - return "", err - } - - return Hash(stdout), nil -} - -func (repo *GitRepo) AllClocks() (map[string]lamport.Clock, error) { - repo.clocksMutex.Lock() - defer repo.clocksMutex.Unlock() - - result := make(map[string]lamport.Clock) - - files, err := ioutil.ReadDir(filepath.Join(repo.path, "git-bug", clockPath)) - if os.IsNotExist(err) { - return nil, nil - } - if err != nil { - return nil, err - } - - for _, file := range files { - name := file.Name() - if c, ok := repo.clocks[name]; ok { - result[name] = c - } else { - c, err := lamport.LoadPersistedClock(repo.LocalStorage(), filepath.Join(clockPath, name)) - if err != nil { - return nil, err - } - repo.clocks[name] = c - result[name] = c - } - } - - return result, nil -} - -// GetOrCreateClock return a Lamport clock stored in the Repo. -// If the clock doesn't exist, it's created. -func (repo *GitRepo) GetOrCreateClock(name string) (lamport.Clock, error) { - repo.clocksMutex.Lock() - defer repo.clocksMutex.Unlock() - - c, err := repo.getClock(name) - if err == nil { - return c, nil - } - if err != ErrClockNotExist { - return nil, err - } - - c, err = lamport.NewPersistedClock(repo.LocalStorage(), filepath.Join(clockPath, name)) - if err != nil { - return nil, err - } - - repo.clocks[name] = c - return c, nil -} - -func (repo *GitRepo) getClock(name string) (lamport.Clock, error) { - if c, ok := repo.clocks[name]; ok { - return c, nil - } - - c, err := lamport.LoadPersistedClock(repo.LocalStorage(), filepath.Join(clockPath, name)) - if err == nil { - repo.clocks[name] = c - return c, nil - } - if err == lamport.ErrClockNotExist { - return nil, ErrClockNotExist - } - return nil, err -} - -// Increment is equivalent to c = GetOrCreateClock(name) + c.Increment() -func (repo *GitRepo) Increment(name string) (lamport.Time, error) { - c, err := repo.GetOrCreateClock(name) - if err != nil { - return lamport.Time(0), err - } - return c.Increment() -} - -// Witness is equivalent to c = GetOrCreateClock(name) + c.Witness(time) -func (repo *GitRepo) Witness(name string, time lamport.Time) error { - c, err := repo.GetOrCreateClock(name) - if err != nil { - return err - } - return c.Witness(time) -} - -// AddRemote add a new remote to the repository -// Not in the interface because it's only used for testing -func (repo *GitRepo) AddRemote(name string, url string) error { - _, err := repo.runGitCommand("remote", "add", name, url) - - return err -} - -// GetLocalRemote return the URL to use to add this repo as a local remote -func (repo *GitRepo) GetLocalRemote() string { - return repo.path -} - -// EraseFromDisk delete this repository entirely from the disk -func (repo *GitRepo) EraseFromDisk() error { - err := repo.Close() - if err != nil { - return err - } - - path := filepath.Clean(strings.TrimSuffix(repo.path, string(filepath.Separator)+".git")) - - // fmt.Println("Cleaning repo:", path) - return os.RemoveAll(path) -} diff --git a/repository/git_cli.go b/repository/git_cli.go deleted file mode 100644 index 085b1cda..00000000 --- a/repository/git_cli.go +++ /dev/null @@ -1,56 +0,0 @@ -package repository - -import ( - "bytes" - "fmt" - "io" - "os/exec" - "strings" -) - -// gitCli is a helper to launch CLI git commands -type gitCli struct { - path string -} - -// Run the given git command with the given I/O reader/writers, returning an error if it fails. -func (cli gitCli) runGitCommandWithIO(stdin io.Reader, stdout, stderr io.Writer, args ...string) error { - // make sure that the working directory for the command - // always exist, in particular when running "git init". - path := strings.TrimSuffix(cli.path, ".git") - - // fmt.Printf("[%s] Running git %s\n", path, strings.Join(args, " ")) - - cmd := exec.Command("git", args...) - cmd.Dir = path - cmd.Stdin = stdin - cmd.Stdout = stdout - cmd.Stderr = stderr - - return cmd.Run() -} - -// Run the given git command and return its stdout, or an error if the command fails. -func (cli gitCli) runGitCommandRaw(stdin io.Reader, args ...string) (string, string, error) { - var stdout bytes.Buffer - var stderr bytes.Buffer - err := cli.runGitCommandWithIO(stdin, &stdout, &stderr, args...) - return strings.TrimSpace(stdout.String()), strings.TrimSpace(stderr.String()), err -} - -// Run the given git command and return its stdout, or an error if the command fails. -func (cli gitCli) runGitCommandWithStdin(stdin io.Reader, args ...string) (string, error) { - stdout, stderr, err := cli.runGitCommandRaw(stdin, args...) - if err != nil { - if stderr == "" { - stderr = "Error running git command: " + strings.Join(args, " ") - } - err = fmt.Errorf(stderr) - } - return stdout, err -} - -// Run the given git command and return its stdout, or an error if the command fails. -func (cli gitCli) runGitCommand(args ...string) (string, error) { - return cli.runGitCommandWithStdin(nil, args...) -} diff --git a/repository/git_config.go b/repository/git_config.go deleted file mode 100644 index b46cc69b..00000000 --- a/repository/git_config.go +++ /dev/null @@ -1,221 +0,0 @@ -package repository - -import ( - "fmt" - "regexp" - "strconv" - "strings" - "time" - - "github.com/blang/semver" - "github.com/pkg/errors" -) - -var _ Config = &gitConfig{} - -type gitConfig struct { - cli gitCli - localityFlag string -} - -func newGitConfig(cli gitCli, global bool) *gitConfig { - localityFlag := "--local" - if global { - localityFlag = "--global" - } - return &gitConfig{ - cli: cli, - localityFlag: localityFlag, - } -} - -// StoreString store a single key/value pair in the config of the repo -func (gc *gitConfig) StoreString(key string, value string) error { - _, err := gc.cli.runGitCommand("config", gc.localityFlag, "--replace-all", key, value) - return err -} - -func (gc *gitConfig) StoreBool(key string, value bool) error { - return gc.StoreString(key, strconv.FormatBool(value)) -} - -func (gc *gitConfig) StoreTimestamp(key string, value time.Time) error { - return gc.StoreString(key, strconv.Itoa(int(value.Unix()))) -} - -// ReadAll read all key/value pair matching the key prefix -func (gc *gitConfig) ReadAll(keyPrefix string) (map[string]string, error) { - stdout, err := gc.cli.runGitCommand("config", gc.localityFlag, "--includes", "--get-regexp", keyPrefix) - - // / \ - // / ! \ - // ------- - // - // There can be a legitimate error here, but I see no portable way to - // distinguish them from the git error that say "no matching value exist" - if err != nil { - return nil, nil - } - - lines := strings.Split(stdout, "\n") - - result := make(map[string]string, len(lines)) - - for _, line := range lines { - if strings.TrimSpace(line) == "" { - continue - } - - parts := strings.SplitN(line, " ", 2) - result[parts[0]] = parts[1] - } - - return result, nil -} - -func (gc *gitConfig) ReadString(key string) (string, error) { - stdout, err := gc.cli.runGitCommand("config", gc.localityFlag, "--includes", "--get-all", key) - - // / \ - // / ! \ - // ------- - // - // There can be a legitimate error here, but I see no portable way to - // distinguish them from the git error that say "no matching value exist" - if err != nil { - return "", ErrNoConfigEntry - } - - lines := strings.Split(stdout, "\n") - - if len(lines) == 0 { - return "", ErrNoConfigEntry - } - if len(lines) > 1 { - return "", ErrMultipleConfigEntry - } - - return lines[0], nil -} - -func (gc *gitConfig) ReadBool(key string) (bool, error) { - val, err := gc.ReadString(key) - if err != nil { - return false, err - } - - return strconv.ParseBool(val) -} - -func (gc *gitConfig) ReadTimestamp(key string) (time.Time, error) { - value, err := gc.ReadString(key) - if err != nil { - return time.Time{}, err - } - return ParseTimestamp(value) -} - -func (gc *gitConfig) rmSection(keyPrefix string) error { - _, err := gc.cli.runGitCommand("config", gc.localityFlag, "--remove-section", keyPrefix) - return err -} - -func (gc *gitConfig) unsetAll(keyPrefix string) error { - _, err := gc.cli.runGitCommand("config", gc.localityFlag, "--unset-all", keyPrefix) - return err -} - -// return keyPrefix section -// example: sectionFromKey(a.b.c.d) return a.b.c -func sectionFromKey(keyPrefix string) string { - s := strings.Split(keyPrefix, ".") - if len(s) == 1 { - return keyPrefix - } - - return strings.Join(s[:len(s)-1], ".") -} - -// rmConfigs with git version lesser than 2.18 -func (gc *gitConfig) rmConfigsGitVersionLT218(keyPrefix string) error { - // try to remove key/value pair by key - err := gc.unsetAll(keyPrefix) - if err != nil { - return gc.rmSection(keyPrefix) - } - - m, err := gc.ReadAll(sectionFromKey(keyPrefix)) - if err != nil { - return err - } - - // if section doesn't have any left key/value remove the section - if len(m) == 0 { - return gc.rmSection(sectionFromKey(keyPrefix)) - } - - return nil -} - -// RmConfigs remove all key/value pair matching the key prefix -func (gc *gitConfig) RemoveAll(keyPrefix string) error { - // starting from git 2.18.0 sections are automatically deleted when the last existing - // key/value is removed. Before 2.18.0 we should remove the section - // see https://github.com/git/git/blob/master/Documentation/RelNotes/2.18.0.txt#L379 - lt218, err := gc.gitVersionLT218() - if err != nil { - return errors.Wrap(err, "getting git version") - } - - if lt218 { - return gc.rmConfigsGitVersionLT218(keyPrefix) - } - - err = gc.unsetAll(keyPrefix) - if err != nil { - return gc.rmSection(keyPrefix) - } - - return nil -} - -func (gc *gitConfig) gitVersion() (*semver.Version, error) { - versionOut, err := gc.cli.runGitCommand("version") - if err != nil { - return nil, err - } - return parseGitVersion(versionOut) -} - -func parseGitVersion(versionOut string) (*semver.Version, error) { - // extract the version and truncate potential bad parts - // ex: 2.23.0.rc1 instead of 2.23.0-rc1 - r := regexp.MustCompile(`(\d+\.){1,2}\d+`) - - extracted := r.FindString(versionOut) - if extracted == "" { - return nil, fmt.Errorf("unreadable git version %s", versionOut) - } - - version, err := semver.Make(extracted) - if err != nil { - return nil, err - } - - return &version, nil -} - -func (gc *gitConfig) gitVersionLT218() (bool, error) { - version, err := gc.gitVersion() - if err != nil { - return false, err - } - - version218string := "2.18.0" - gitVersion218, err := semver.Make(version218string) - if err != nil { - return false, err - } - - return version.LT(gitVersion218), nil -} diff --git a/repository/git_test.go b/repository/git_test.go deleted file mode 100644 index 6603e339..00000000 --- a/repository/git_test.go +++ /dev/null @@ -1,6 +0,0 @@ -// Package repository contains helper methods for working with the Git repo. -package repository - -// func TestGitRepo(t *testing.T) { -// RepoTest(t, CreateTestRepo, CleanupTestRepos) -// } diff --git a/repository/git_testing.go b/repository/git_testing.go deleted file mode 100644 index 2168d53e..00000000 --- a/repository/git_testing.go +++ /dev/null @@ -1,72 +0,0 @@ -package repository - -import ( - "io/ioutil" - "log" - - "github.com/99designs/keyring" -) - -// This is intended for testing only - -func CreateTestRepo(bare bool) TestedRepo { - dir, err := ioutil.TempDir("", "") - if err != nil { - log.Fatal(err) - } - - var creator func(string) (*GitRepo, error) - - if bare { - creator = InitBareGitRepo - } else { - creator = InitGitRepo - } - - repo, err := creator(dir) - if err != nil { - log.Fatal(err) - } - - config := repo.LocalConfig() - if err := config.StoreString("user.name", "testuser"); err != nil { - log.Fatal("failed to set user.name for test repository: ", err) - } - if err := config.StoreString("user.email", "testuser@example.com"); err != nil { - log.Fatal("failed to set user.email for test repository: ", err) - } - - // make sure we use a mock keyring for testing to not interact with the global system - return &replaceKeyring{ - TestedRepo: repo, - keyring: keyring.NewArrayKeyring(nil), - } -} - -func SetupReposAndRemote() (repoA, repoB, remote TestedRepo) { - repoA = CreateGoGitTestRepo(false) - repoB = CreateGoGitTestRepo(false) - remote = CreateGoGitTestRepo(true) - - err := repoA.AddRemote("origin", remote.GetLocalRemote()) - if err != nil { - log.Fatal(err) - } - - err = repoB.AddRemote("origin", remote.GetLocalRemote()) - if err != nil { - log.Fatal(err) - } - - return repoA, repoB, remote -} - -// replaceKeyring allow to replace the Keyring of the underlying repo -type replaceKeyring struct { - TestedRepo - keyring Keyring -} - -func (rk replaceKeyring) Keyring() Keyring { - return rk.keyring -} diff --git a/repository/gogit.go b/repository/gogit.go index 64ccb773..d6eb8621 100644 --- a/repository/gogit.go +++ b/repository/gogit.go @@ -20,6 +20,7 @@ import ( "github.com/go-git/go-git/v5/plumbing" "github.com/go-git/go-git/v5/plumbing/filemode" "github.com/go-git/go-git/v5/plumbing/object" + "golang.org/x/crypto/openpgp" "github.com/MichaelMure/git-bug/util/lamport" ) @@ -521,12 +522,13 @@ func (repo *GoGitRepo) ReadTree(hash Hash) ([]TreeEntry, error) { } // StoreCommit will store a Git commit with the given Git tree -func (repo *GoGitRepo) StoreCommit(treeHash Hash) (Hash, error) { - return repo.StoreCommitWithParent(treeHash, "") +func (repo *GoGitRepo) StoreCommit(treeHash Hash, parents ...Hash) (Hash, error) { + return repo.StoreSignedCommit(treeHash, nil, parents...) } -// StoreCommit will store a Git commit with the given Git tree -func (repo *GoGitRepo) StoreCommitWithParent(treeHash Hash, parent Hash) (Hash, error) { +// StoreCommit will store a Git commit with the given Git tree. If signKey is not nil, the commit +// will be signed accordingly. +func (repo *GoGitRepo) StoreSignedCommit(treeHash Hash, signKey *openpgp.Entity, parents ...Hash) (Hash, error) { cfg, err := repo.r.Config() if err != nil { return "", err @@ -547,8 +549,28 @@ func (repo *GoGitRepo) StoreCommitWithParent(treeHash Hash, parent Hash) (Hash, TreeHash: plumbing.NewHash(treeHash.String()), } - if parent != "" { - commit.ParentHashes = []plumbing.Hash{plumbing.NewHash(parent.String())} + for _, parent := range parents { + commit.ParentHashes = append(commit.ParentHashes, plumbing.NewHash(parent.String())) + } + + // Compute the signature if needed + if signKey != nil { + // first get the serialized commit + encoded := &plumbing.MemoryObject{} + if err := commit.Encode(encoded); err != nil { + return "", err + } + r, err := encoded.Reader() + if err != nil { + return "", err + } + + // sign the data + var sig bytes.Buffer + if err := openpgp.ArmoredDetachSign(&sig, signKey, r, nil); err != nil { + return "", err + } + commit.PGPSignature = sig.String() } obj := repo.r.Storer.NewEncodedObject() @@ -608,6 +630,13 @@ func (repo *GoGitRepo) UpdateRef(ref string, hash Hash) error { return repo.r.Storer.SetReference(plumbing.NewHashReference(plumbing.ReferenceName(ref), plumbing.NewHash(hash.String()))) } +// MergeRef merge other into ref and update the reference +// If the update is not fast-forward, the callback treeHashFn will be called for the caller to generate +// the Tree to store in the merge commit. +func (repo *GoGitRepo) MergeRef(ref string, otherRef string, treeHashFn func() Hash) error { + return nonNativeMerge(repo, ref, otherRef, treeHashFn) +} + // RemoveRef will remove a Git reference func (repo *GoGitRepo) RemoveRef(ref string) error { return repo.r.Storer.RemoveReference(plumbing.ReferenceName(ref)) @@ -657,38 +686,16 @@ func (repo *GoGitRepo) CopyRef(source string, dest string) error { // ListCommits will return the list of tree hashes of a ref, in chronological order func (repo *GoGitRepo) ListCommits(ref string) ([]Hash, error) { - r, err := repo.r.Reference(plumbing.ReferenceName(ref), false) - if err != nil { - return nil, err - } + return nonNativeListCommits(repo, ref) +} - commit, err := repo.r.CommitObject(r.Hash()) +func (repo *GoGitRepo) ReadCommit(hash Hash) (Commit, error) { + encoded, err := repo.r.Storer.EncodedObject(plumbing.CommitObject, plumbing.NewHash(hash.String())) if err != nil { - return nil, err - } - hashes := []Hash{Hash(commit.Hash.String())} - - for { - commit, err = commit.Parent(0) - if err == object.ErrParentNotFound { - break - } - if err != nil { - return nil, err - } - - if commit.NumParents() > 1 { - return nil, fmt.Errorf("multiple parents") - } - - hashes = append([]Hash{Hash(commit.Hash.String())}, hashes...) + return Commit{}, err } - return hashes, nil -} - -func (repo *GoGitRepo) ReadCommit(hash Hash) (Commit, error) { - commit, err := repo.r.CommitObject(plumbing.NewHash(hash.String())) + commit, err := object.DecodeCommit(repo.r.Storer, encoded) if err != nil { return Commit{}, err } @@ -698,12 +705,25 @@ func (repo *GoGitRepo) ReadCommit(hash Hash) (Commit, error) { parents[i] = Hash(parentHash.String()) } - return Commit{ + result := Commit{ Hash: hash, Parents: parents, TreeHash: Hash(commit.TreeHash.String()), - }, nil + } + + if commit.PGPSignature != "" { + result.SignedData, err = encoded.Reader() + if err != nil { + return Commit{}, err + } + result.Signature, err = deArmorSignature(strings.NewReader(commit.PGPSignature)) + if err != nil { + return Commit{}, err + } + } + + return result, nil } func (repo *GoGitRepo) AllClocks() (map[string]lamport.Clock, error) { diff --git a/repository/gogit_testing.go b/repository/gogit_testing.go index a8bff41e..cad776b3 100644 --- a/repository/gogit_testing.go +++ b/repository/gogit_testing.go @@ -3,6 +3,8 @@ package repository import ( "io/ioutil" "log" + + "github.com/99designs/keyring" ) // This is intended for testing only @@ -34,7 +36,11 @@ func CreateGoGitTestRepo(bare bool) TestedRepo { log.Fatal("failed to set user.email for test repository: ", err) } - return repo + // make sure we use a mock keyring for testing to not interact with the global system + return &replaceKeyring{ + TestedRepo: repo, + keyring: keyring.NewArrayKeyring(nil), + } } func SetupGoGitReposAndRemote() (repoA, repoB, remote TestedRepo) { diff --git a/repository/keyring.go b/repository/keyring.go index 4cb3c9ff..64365c39 100644 --- a/repository/keyring.go +++ b/repository/keyring.go @@ -48,3 +48,13 @@ func defaultKeyring() (Keyring, error) { }, }) } + +// replaceKeyring allow to replace the Keyring of the underlying repo +type replaceKeyring struct { + TestedRepo + keyring Keyring +} + +func (rk replaceKeyring) Keyring() Keyring { + return rk.keyring +} diff --git a/repository/mock_repo.go b/repository/mock_repo.go index 227e0f2c..095ad61c 100644 --- a/repository/mock_repo.go +++ b/repository/mock_repo.go @@ -1,6 +1,7 @@ package repository import ( + "bytes" "crypto/sha1" "fmt" "strings" @@ -10,6 +11,7 @@ import ( "github.com/blevesearch/bleve" "github.com/go-git/go-billy/v5" "github.com/go-git/go-billy/v5/memfs" + "golang.org/x/crypto/openpgp" "github.com/MichaelMure/git-bug/util/lamport" ) @@ -180,6 +182,7 @@ var _ RepoData = &mockRepoData{} type commit struct { treeHash Hash parents []Hash + sig string } type mockRepoData struct { @@ -198,12 +201,12 @@ func NewMockRepoData() *mockRepoData { } } -// PushRefs push git refs to a remote -func (r *mockRepoData) PushRefs(remote string, refSpec string) (string, error) { +func (r *mockRepoData) FetchRefs(remote string, refSpec string) (string, error) { return "", nil } -func (r *mockRepoData) FetchRefs(remote string, refSpec string) (string, error) { +// PushRefs push git refs to a remote +func (r *mockRepoData) PushRefs(remote string, refSpec string) (string, error) { return "", nil } @@ -216,7 +219,6 @@ func (r *mockRepoData) StoreData(data []byte) (Hash, error) { func (r *mockRepoData) ReadData(hash Hash) ([]byte, error) { data, ok := r.blobs[hash] - if !ok { return nil, fmt.Errorf("unknown hash") } @@ -233,25 +235,86 @@ func (r *mockRepoData) StoreTree(entries []TreeEntry) (Hash, error) { return hash, nil } -func (r *mockRepoData) StoreCommit(treeHash Hash) (Hash, error) { - rawHash := sha1.Sum([]byte(treeHash)) - hash := Hash(fmt.Sprintf("%x", rawHash)) - r.commits[hash] = commit{ - treeHash: treeHash, +func (r *mockRepoData) ReadTree(hash Hash) ([]TreeEntry, error) { + var data string + + data, ok := r.trees[hash] + + if !ok { + // Git will understand a commit hash to reach a tree + commit, ok := r.commits[hash] + + if !ok { + return nil, fmt.Errorf("unknown hash") + } + + data, ok = r.trees[commit.treeHash] + + if !ok { + return nil, fmt.Errorf("unknown hash") + } } - return hash, nil + + return readTreeEntries(data) +} + +func (r *mockRepoData) StoreCommit(treeHash Hash, parents ...Hash) (Hash, error) { + return r.StoreSignedCommit(treeHash, nil, parents...) } -func (r *mockRepoData) StoreCommitWithParent(treeHash Hash, parent Hash) (Hash, error) { - rawHash := sha1.Sum([]byte(treeHash + parent)) +func (r *mockRepoData) StoreSignedCommit(treeHash Hash, signKey *openpgp.Entity, parents ...Hash) (Hash, error) { + hasher := sha1.New() + hasher.Write([]byte(treeHash)) + for _, parent := range parents { + hasher.Write([]byte(parent)) + } + rawHash := hasher.Sum(nil) hash := Hash(fmt.Sprintf("%x", rawHash)) - r.commits[hash] = commit{ + c := commit{ treeHash: treeHash, - parents: []Hash{parent}, + parents: parents, + } + if signKey != nil { + // unlike go-git, we only sign the tree hash for simplicity instead of all the fields (parents ...) + var sig bytes.Buffer + if err := openpgp.DetachSign(&sig, signKey, strings.NewReader(string(treeHash)), nil); err != nil { + return "", err + } + c.sig = sig.String() } + r.commits[hash] = c return hash, nil } +func (r *mockRepoData) ReadCommit(hash Hash) (Commit, error) { + c, ok := r.commits[hash] + if !ok { + return Commit{}, fmt.Errorf("unknown commit") + } + + result := Commit{ + Hash: hash, + Parents: c.parents, + TreeHash: c.treeHash, + } + + if c.sig != "" { + result.SignedData = strings.NewReader(string(c.treeHash)) + result.Signature = strings.NewReader(c.sig) + } + + return result, nil +} + +func (r *mockRepoData) GetTreeHash(commit Hash) (Hash, error) { + c, ok := r.commits[commit] + if !ok { + return "", fmt.Errorf("unknown commit") + } + + return c.treeHash, nil +} + func (r *mockRepoData) ResolveRef(ref string) (Hash, error) { h, ok := r.refs[ref] if !ok { @@ -270,22 +333,6 @@ func (r *mockRepoData) RemoveRef(ref string) error { return nil } -func (r *mockRepoData) RefExist(ref string) (bool, error) { - _, exist := r.refs[ref] - return exist, nil -} - -func (r *mockRepoData) CopyRef(source string, dest string) error { - hash, exist := r.refs[source] - - if !exist { - return fmt.Errorf("Unknown ref") - } - - r.refs[dest] = hash - return nil -} - func (r *mockRepoData) ListRefs(refPrefix string) ([]string, error) { var keys []string @@ -298,63 +345,20 @@ func (r *mockRepoData) ListRefs(refPrefix string) ([]string, error) { return keys, nil } -func (r *mockRepoData) ListCommits(ref string) ([]Hash, error) { - var hashes []Hash - - hash := r.refs[ref] - - for { - commit, ok := r.commits[hash] - - if !ok { - break - } - - hashes = append([]Hash{hash}, hashes...) - - if len(commit.parents) == 0 { - break - } - hash = commit.parents[0] - } - - return hashes, nil -} - -func (r *mockRepoData) ReadCommit(hash Hash) (Commit, error) { - c, ok := r.commits[hash] - if !ok { - return Commit{}, fmt.Errorf("unknown commit") - } - - return Commit{ - Hash: hash, - Parents: c.parents, - TreeHash: c.treeHash, - }, nil +func (r *mockRepoData) RefExist(ref string) (bool, error) { + _, exist := r.refs[ref] + return exist, nil } -func (r *mockRepoData) ReadTree(hash Hash) ([]TreeEntry, error) { - var data string - - data, ok := r.trees[hash] - - if !ok { - // Git will understand a commit hash to reach a tree - commit, ok := r.commits[hash] - - if !ok { - return nil, fmt.Errorf("unknown hash") - } - - data, ok = r.trees[commit.treeHash] +func (r *mockRepoData) CopyRef(source string, dest string) error { + hash, exist := r.refs[source] - if !ok { - return nil, fmt.Errorf("unknown hash") - } + if !exist { + return fmt.Errorf("Unknown ref") } - return readTreeEntries(data) + r.refs[dest] = hash + return nil } func (r *mockRepoData) FindCommonAncestor(hash1 Hash, hash2 Hash) (Hash, error) { @@ -392,13 +396,8 @@ func (r *mockRepoData) FindCommonAncestor(hash1 Hash, hash2 Hash) (Hash, error) } } -func (r *mockRepoData) GetTreeHash(commit Hash) (Hash, error) { - c, ok := r.commits[commit] - if !ok { - return "", fmt.Errorf("unknown commit") - } - - return c.treeHash, nil +func (r *mockRepoData) ListCommits(ref string) ([]Hash, error) { + return nonNativeListCommits(r, ref) } var _ RepoClock = &mockRepoClock{} diff --git a/repository/mock_repo_test.go b/repository/mock_repo_test.go index dec09380..12851a80 100644 --- a/repository/mock_repo_test.go +++ b/repository/mock_repo_test.go @@ -1,6 +1,8 @@ package repository -import "testing" +import ( + "testing" +) func TestMockRepo(t *testing.T) { creator := func(bare bool) TestedRepo { return NewMockRepo() } diff --git a/repository/repo.go b/repository/repo.go index afd8ff77..d7afa983 100644 --- a/repository/repo.go +++ b/repository/repo.go @@ -3,15 +3,17 @@ package repository import ( "errors" + "io" "github.com/blevesearch/bleve" "github.com/go-git/go-billy/v5" + "golang.org/x/crypto/openpgp" "github.com/MichaelMure/git-bug/util/lamport" ) var ( - // ErrNotARepo is the error returned when the git repo root wan't be found + // ErrNotARepo is the error returned when the git repo root can't be found ErrNotARepo = errors.New("not a git repository") // ErrClockNotExist is the error returned when a clock can't be found ErrClockNotExist = errors.New("clock doesn't exist") @@ -89,9 +91,11 @@ type RepoBleve interface { } type Commit struct { - Hash Hash - Parents []Hash - TreeHash Hash + Hash Hash + Parents []Hash // hashes of the parents, if any + TreeHash Hash // hash of the git Tree + SignedData io.Reader // if signed, reader for the signed data (likely, the serialized commit) + Signature io.Reader // if signed, reader for the (non-armored) signature } // RepoData give access to the git data storage @@ -116,21 +120,29 @@ type RepoData interface { ReadTree(hash Hash) ([]TreeEntry, error) // StoreCommit will store a Git commit with the given Git tree - StoreCommit(treeHash Hash) (Hash, error) + StoreCommit(treeHash Hash, parents ...Hash) (Hash, error) - // StoreCommit will store a Git commit with the given Git tree - StoreCommitWithParent(treeHash Hash, parent Hash) (Hash, error) + // StoreCommit will store a Git commit with the given Git tree. If signKey is not nil, the commit + // will be signed accordingly. + StoreSignedCommit(treeHash Hash, signKey *openpgp.Entity, parents ...Hash) (Hash, error) + // ReadCommit read a Git commit and returns some of its characteristic ReadCommit(hash Hash) (Commit, error) // GetTreeHash return the git tree hash referenced in a commit GetTreeHash(commit Hash) (Hash, error) + // ResolveRef returns the hash of the target commit of the given ref ResolveRef(ref string) (Hash, error) // UpdateRef will create or update a Git reference UpdateRef(ref string, hash Hash) error + // // MergeRef merge other into ref and update the reference + // // If the update is not fast-forward, the callback treeHashFn will be called for the caller to generate + // // the Tree to store in the merge commit. + // MergeRef(ref string, otherRef string, treeHashFn func() Hash) error + // RemoveRef will remove a Git reference RemoveRef(ref string) error @@ -148,7 +160,6 @@ type RepoData interface { FindCommonAncestor(commit1 Hash, commit2 Hash) (Hash, error) // ListCommits will return the list of tree hashes of a ref, in chronological order - // Deprecated ListCommits(ref string) ([]Hash, error) } diff --git a/repository/repo_testing.go b/repository/repo_testing.go index 1d3a3155..4a5c48bb 100644 --- a/repository/repo_testing.go +++ b/repository/repo_testing.go @@ -135,7 +135,8 @@ func RepoDataTest(t *testing.T, repo RepoData) { require.NoError(t, err) require.Equal(t, treeHash1, treeHash1Read) - commit2, err := repo.StoreCommitWithParent(treeHash2, commit1) + // commit with a parent + commit2, err := repo.StoreCommit(treeHash2, commit1) require.NoError(t, err) require.True(t, commit2.IsValid()) @@ -187,7 +188,7 @@ func RepoDataTest(t *testing.T, repo RepoData) { // Graph - commit3, err := repo.StoreCommitWithParent(treeHash1, commit1) + commit3, err := repo.StoreCommit(treeHash1, commit1) require.NoError(t, err) ancestorHash, err := repo.FindCommonAncestor(commit2, commit3) @@ -237,3 +238,22 @@ func randomData() []byte { } return b } + +func makeCommit(t *testing.T, repo RepoData, parents ...Hash) Hash { + blobHash, err := repo.StoreData(randomData()) + require.NoError(t, err) + + treeHash, err := repo.StoreTree([]TreeEntry{ + { + ObjectType: Blob, + Hash: blobHash, + Name: "foo", + }, + }) + require.NoError(t, err) + + commitHash, err := repo.StoreCommit(treeHash, parents...) + require.NoError(t, err) + + return commitHash +} diff --git a/util/lamport/clock_testing.go b/util/lamport/clock_testing.go index 4bf6d2bf..de66c5c9 100644 --- a/util/lamport/clock_testing.go +++ b/util/lamport/clock_testing.go @@ -14,11 +14,11 @@ func testClock(t *testing.T, c Clock) { assert.Equal(t, Time(2), val) assert.Equal(t, Time(2), c.Time()) - err = c.Witness(41) + err = c.Witness(42) assert.NoError(t, err) assert.Equal(t, Time(42), c.Time()) - err = c.Witness(41) + err = c.Witness(42) assert.NoError(t, err) assert.Equal(t, Time(42), c.Time()) diff --git a/util/lamport/mem_clock.go b/util/lamport/mem_clock.go index f113b501..d824d834 100644 --- a/util/lamport/mem_clock.go +++ b/util/lamport/mem_clock.go @@ -25,6 +25,14 @@ */ +// Note: this code originally originate from Hashicorp's Serf but has been changed since to fit git-bug's need. + +// Note: this Lamport clock implementation is different than the algorithms you can find, notably Wikipedia or the +// original Serf implementation. The reason is lie to what constitute an event in this distributed system. +// Commonly, events happen when messages are sent or received, whereas in git-bug events happen when some data is +// written, but *not* when read. This is why Witness set the time to the max seen value instead of max seen value +1. +// See https://cs.stackexchange.com/a/133730/129795 + package lamport import ( @@ -72,12 +80,12 @@ WITNESS: // If the other value is old, we do not need to do anything cur := atomic.LoadUint64(&mc.counter) other := uint64(v) - if other < cur { + if other <= cur { return nil } // Ensure that our local clock is at least one ahead. - if !atomic.CompareAndSwapUint64(&mc.counter, cur, other+1) { + if !atomic.CompareAndSwapUint64(&mc.counter, cur, other) { // CAS: CompareAndSwap // The CAS failed, so we just retry. Eventually our CAS should // succeed or a future witness will pass us by and our witness |