1// Package dag contains the base common code to define an entity stored
  2// in a chain of git objects, supporting actions like Push, Pull and Merge.
  3package dag
  4
  5import (
  6	"encoding/json"
  7	"fmt"
  8	"sort"
  9
 10	"github.com/pkg/errors"
 11
 12	"github.com/MichaelMure/git-bug/entity"
 13	"github.com/MichaelMure/git-bug/identity"
 14	"github.com/MichaelMure/git-bug/repository"
 15	"github.com/MichaelMure/git-bug/util/lamport"
 16)
 17
 18const refsPattern = "refs/%s/%s"
 19const creationClockPattern = "%s-create"
 20const editClockPattern = "%s-edit"
 21
 22// Definition hold the details defining one specialization of an Entity.
 23type Definition struct {
 24	// the name of the entity (bug, pull-request, ...), for human consumption
 25	Typename string
 26	// the Namespace in git references (bugs, prs, ...)
 27	Namespace string
 28	// a function decoding a JSON message into an Operation
 29	OperationUnmarshaler func(author identity.Interface, raw json.RawMessage) (Operation, error)
 30	// the expected format version number, that can be used for data migration/upgrade
 31	FormatVersion uint
 32}
 33
 34// Entity is a data structure stored in a chain of git objects, supporting actions like Push, Pull and Merge.
 35type Entity struct {
 36	// A Lamport clock is a logical clock that allow to order event
 37	// inside a distributed system.
 38	// It must be the first field in this struct due to https://github.com/golang/go/issues/36606
 39	createTime lamport.Time
 40	editTime   lamport.Time
 41
 42	Definition
 43
 44	// operations that are already stored in the repository
 45	ops []Operation
 46	// operations not yet stored in the repository
 47	staging []Operation
 48
 49	lastCommit repository.Hash
 50}
 51
 52// New create an empty Entity
 53func New(definition Definition) *Entity {
 54	return &Entity{
 55		Definition: definition,
 56	}
 57}
 58
 59// Read will read and decode a stored local Entity from a repository
 60func Read(def Definition, repo repository.ClockedRepo, resolver identity.Resolver, id entity.Id) (*Entity, error) {
 61	if err := id.Validate(); err != nil {
 62		return nil, errors.Wrap(err, "invalid id")
 63	}
 64
 65	ref := fmt.Sprintf("refs/%s/%s", def.Namespace, id.String())
 66
 67	return read(def, repo, resolver, ref)
 68}
 69
 70// readRemote will read and decode a stored remote Entity from a repository
 71func readRemote(def Definition, repo repository.ClockedRepo, resolver identity.Resolver, remote string, id entity.Id) (*Entity, error) {
 72	if err := id.Validate(); err != nil {
 73		return nil, errors.Wrap(err, "invalid id")
 74	}
 75
 76	ref := fmt.Sprintf("refs/remotes/%s/%s/%s", def.Namespace, remote, id.String())
 77
 78	return read(def, repo, resolver, ref)
 79}
 80
 81// read fetch from git and decode an Entity at an arbitrary git reference.
 82func read(def Definition, repo repository.ClockedRepo, resolver identity.Resolver, ref string) (*Entity, error) {
 83	rootHash, err := repo.ResolveRef(ref)
 84	if err != nil {
 85		return nil, err
 86	}
 87
 88	// Perform a breadth-first search to get a topological order of the DAG where we discover the
 89	// parents commit and go back in time up to the chronological root
 90
 91	queue := make([]repository.Hash, 0, 32)
 92	visited := make(map[repository.Hash]struct{})
 93	BFSOrder := make([]repository.Commit, 0, 32)
 94
 95	queue = append(queue, rootHash)
 96	visited[rootHash] = struct{}{}
 97
 98	for len(queue) > 0 {
 99		// pop
100		hash := queue[0]
101		queue = queue[1:]
102
103		commit, err := repo.ReadCommit(hash)
104		if err != nil {
105			return nil, err
106		}
107
108		BFSOrder = append(BFSOrder, commit)
109
110		for _, parent := range commit.Parents {
111			if _, ok := visited[parent]; !ok {
112				queue = append(queue, parent)
113				// mark as visited
114				visited[parent] = struct{}{}
115			}
116		}
117	}
118
119	// Now, we can reverse this topological order and read the commits in an order where
120	// we are sure to have read all the chronological ancestors when we read a commit.
121
122	// Next step is to:
123	// 1) read the operationPacks
124	// 2) make sure that clocks causality respect the DAG topology.
125
126	oppMap := make(map[repository.Hash]*operationPack)
127	var opsCount int
128
129	for i := len(BFSOrder) - 1; i >= 0; i-- {
130		commit := BFSOrder[i]
131		isFirstCommit := i == len(BFSOrder)-1
132		isMerge := len(commit.Parents) > 1
133
134		// Verify DAG structure: single chronological root, so only the root
135		// can have no parents. Said otherwise, the DAG need to have exactly
136		// one leaf.
137		if !isFirstCommit && len(commit.Parents) == 0 {
138			return nil, fmt.Errorf("multiple leafs in the entity DAG")
139		}
140
141		opp, err := readOperationPack(def, repo, resolver, commit)
142		if err != nil {
143			return nil, err
144		}
145
146		err = opp.Validate()
147		if err != nil {
148			return nil, err
149		}
150
151		if isMerge && len(opp.Operations) > 0 {
152			return nil, fmt.Errorf("merge commit cannot have operations")
153		}
154
155		// Check that the create lamport clock is set (not checked in Validate() as it's optional)
156		if isFirstCommit && opp.CreateTime <= 0 {
157			return nil, fmt.Errorf("creation lamport time not set")
158		}
159
160		// make sure that the lamport clocks causality match the DAG topology
161		for _, parentHash := range commit.Parents {
162			parentPack, ok := oppMap[parentHash]
163			if !ok {
164				panic("DFS failed")
165			}
166
167			if parentPack.EditTime >= opp.EditTime {
168				return nil, fmt.Errorf("lamport clock ordering doesn't match the DAG")
169			}
170
171			// to avoid an attack where clocks are pushed toward the uint64 rollover, make sure
172			// that the clocks don't jump too far in the future
173			// we ignore merge commits here to allow merging after a loooong time without breaking anything,
174			// as long as there is one valid chain of small hops, it's fine.
175			if !isMerge && opp.EditTime-parentPack.EditTime > 1_000_000 {
176				return nil, fmt.Errorf("lamport clock jumping too far in the future, likely an attack")
177			}
178		}
179
180		oppMap[commit.Hash] = opp
181		opsCount += len(opp.Operations)
182	}
183
184	// The clocks are fine, we witness them
185	for _, opp := range oppMap {
186		err = repo.Witness(fmt.Sprintf(creationClockPattern, def.Namespace), opp.CreateTime)
187		if err != nil {
188			return nil, err
189		}
190		err = repo.Witness(fmt.Sprintf(editClockPattern, def.Namespace), opp.EditTime)
191		if err != nil {
192			return nil, err
193		}
194	}
195
196	// Now that we know that the topological order and clocks are fine, we order the operationPacks
197	// based on the logical clocks, entirely ignoring the DAG topology
198
199	oppSlice := make([]*operationPack, 0, len(oppMap))
200	for _, pack := range oppMap {
201		oppSlice = append(oppSlice, pack)
202	}
203	sort.Slice(oppSlice, func(i, j int) bool {
204		// Primary ordering with the EditTime.
205		if oppSlice[i].EditTime != oppSlice[j].EditTime {
206			return oppSlice[i].EditTime < oppSlice[j].EditTime
207		}
208		// We have equal EditTime, which means we have concurrent edition over different machines and we
209		// can't tell which one came first. So, what now? We still need a total ordering and the most stable possible.
210		// As a secondary ordering, we can order based on a hash of the serialized Operations in the
211		// operationPack. It doesn't carry much meaning but it's unbiased and hard to abuse.
212		// This is a lexicographic ordering on the stringified ID.
213		return oppSlice[i].Id() < oppSlice[j].Id()
214	})
215
216	// Now that we ordered the operationPacks, we have the order of the Operations
217
218	ops := make([]Operation, 0, opsCount)
219	var createTime lamport.Time
220	var editTime lamport.Time
221	for _, pack := range oppSlice {
222		for _, operation := range pack.Operations {
223			ops = append(ops, operation)
224		}
225		if pack.CreateTime > createTime {
226			createTime = pack.CreateTime
227		}
228		if pack.EditTime > editTime {
229			editTime = pack.EditTime
230		}
231	}
232
233	return &Entity{
234		Definition: def,
235		ops:        ops,
236		lastCommit: rootHash,
237		createTime: createTime,
238		editTime:   editTime,
239	}, nil
240}
241
242type StreamedEntity struct {
243	Entity *Entity
244	Err    error
245}
246
247// ReadAll read and parse all local Entity
248func ReadAll(def Definition, repo repository.ClockedRepo, resolver identity.Resolver) <-chan StreamedEntity {
249	out := make(chan StreamedEntity)
250
251	go func() {
252		defer close(out)
253
254		refPrefix := fmt.Sprintf("refs/%s/", def.Namespace)
255
256		refs, err := repo.ListRefs(refPrefix)
257		if err != nil {
258			out <- StreamedEntity{Err: err}
259			return
260		}
261
262		for _, ref := range refs {
263			e, err := read(def, repo, resolver, ref)
264
265			if err != nil {
266				out <- StreamedEntity{Err: err}
267				return
268			}
269
270			out <- StreamedEntity{Entity: e}
271		}
272	}()
273
274	return out
275}
276
277// Id return the Entity identifier
278func (e *Entity) Id() entity.Id {
279	// id is the id of the first operation
280	return e.FirstOp().Id()
281}
282
283// Validate check if the Entity data is valid
284func (e *Entity) Validate() error {
285	// non-empty
286	if len(e.ops) == 0 && len(e.staging) == 0 {
287		return fmt.Errorf("entity has no operations")
288	}
289
290	// check if each operations are valid
291	for _, op := range e.ops {
292		if err := op.Validate(); err != nil {
293			return err
294		}
295	}
296
297	// check if staging is valid if needed
298	for _, op := range e.staging {
299		if err := op.Validate(); err != nil {
300			return err
301		}
302	}
303
304	// Check that there is no colliding operation's ID
305	ids := make(map[entity.Id]struct{})
306	for _, op := range e.Operations() {
307		if _, ok := ids[op.Id()]; ok {
308			return fmt.Errorf("id collision: %s", op.Id())
309		}
310		ids[op.Id()] = struct{}{}
311	}
312
313	return nil
314}
315
316// Operations return the ordered operations
317func (e *Entity) Operations() []Operation {
318	return append(e.ops, e.staging...)
319}
320
321// FirstOp lookup for the very first operation of the Entity
322func (e *Entity) FirstOp() Operation {
323	for _, op := range e.ops {
324		return op
325	}
326	for _, op := range e.staging {
327		return op
328	}
329	return nil
330}
331
332// LastOp lookup for the very last operation of the Entity
333func (e *Entity) LastOp() Operation {
334	if len(e.staging) > 0 {
335		return e.staging[len(e.staging)-1]
336	}
337	if len(e.ops) > 0 {
338		return e.ops[len(e.ops)-1]
339	}
340	return nil
341}
342
343// Append add a new Operation to the Entity
344func (e *Entity) Append(op Operation) {
345	e.staging = append(e.staging, op)
346}
347
348// NeedCommit indicate if the in-memory state changed and need to be commit in the repository
349func (e *Entity) NeedCommit() bool {
350	return len(e.staging) > 0
351}
352
353// CommitAsNeeded execute a Commit only if necessary. This function is useful to avoid getting an error if the Entity
354// is already in sync with the repository.
355func (e *Entity) CommitAsNeeded(repo repository.ClockedRepo) error {
356	if e.NeedCommit() {
357		return e.Commit(repo)
358	}
359	return nil
360}
361
362// Commit write the appended operations in the repository
363func (e *Entity) Commit(repo repository.ClockedRepo) error {
364	if !e.NeedCommit() {
365		return fmt.Errorf("can't commit an entity with no pending operation")
366	}
367
368	err := e.Validate()
369	if err != nil {
370		return errors.Wrapf(err, "can't commit a %s with invalid data", e.Definition.Typename)
371	}
372
373	for len(e.staging) > 0 {
374		var author identity.Interface
375		var toCommit []Operation
376
377		// Split into chunks with the same author
378		for len(e.staging) > 0 {
379			op := e.staging[0]
380			if author != nil && op.Author().Id() != author.Id() {
381				break
382			}
383			author = e.staging[0].Author()
384			toCommit = append(toCommit, op)
385			e.staging = e.staging[1:]
386		}
387
388		e.editTime, err = repo.Increment(fmt.Sprintf(editClockPattern, e.Namespace))
389		if err != nil {
390			return err
391		}
392
393		opp := &operationPack{
394			Author:     author,
395			Operations: toCommit,
396			EditTime:   e.editTime,
397		}
398
399		if e.lastCommit == "" {
400			e.createTime, err = repo.Increment(fmt.Sprintf(creationClockPattern, e.Namespace))
401			if err != nil {
402				return err
403			}
404			opp.CreateTime = e.createTime
405		}
406
407		var parentCommit []repository.Hash
408		if e.lastCommit != "" {
409			parentCommit = []repository.Hash{e.lastCommit}
410		}
411
412		commitHash, err := opp.Write(e.Definition, repo, parentCommit...)
413		if err != nil {
414			return err
415		}
416
417		e.lastCommit = commitHash
418		e.ops = append(e.ops, toCommit...)
419	}
420
421	// not strictly necessary but make equality testing easier in tests
422	e.staging = nil
423
424	// Create or update the Git reference for this entity
425	// When pushing later, the remote will ensure that this ref update
426	// is fast-forward, that is no data has been overwritten.
427	ref := fmt.Sprintf(refsPattern, e.Namespace, e.Id().String())
428	return repo.UpdateRef(ref, e.lastCommit)
429}
430
431// CreateLamportTime return the Lamport time of creation
432func (e *Entity) CreateLamportTime() lamport.Time {
433	return e.createTime
434}
435
436// EditLamportTime return the Lamport time of the last edition
437func (e *Entity) EditLamportTime() lamport.Time {
438	return e.editTime
439}