dehub/repo.go

// Package dehub TODO needs package docs
package dehub

import (
	"bytes"
	"encoding/hex"
	"errors"
	"fmt"
	"io"
	"os"
	"path/filepath"
	"strings"

	"dehub.dev/src/dehub.git/fs"

	"gopkg.in/src-d/go-billy.v4"
	"gopkg.in/src-d/go-billy.v4/memfs"
	"gopkg.in/src-d/go-git.v4"
	"gopkg.in/src-d/go-git.v4/plumbing"
	"gopkg.in/src-d/go-git.v4/plumbing/cache"
	"gopkg.in/src-d/go-git.v4/plumbing/format/config"
	"gopkg.in/src-d/go-git.v4/plumbing/object"
	"gopkg.in/src-d/go-git.v4/storage"
	"gopkg.in/src-d/go-git.v4/storage/filesystem"
)

const (
	// DehubDir defines the name of the directory where all dehub-related files are
	// expected to be found.
	DehubDir = ".dehub"
)

var (
	// ConfigPath defines the expected path to the Repo's configuration file.
	ConfigPath = filepath.Join(DehubDir, "config.yml")

	// Main defines the name of the main branch.
	Main = "main"

	// MainRefName defines the reference name of the main branch.
	MainRefName = plumbing.NewBranchReferenceName(Main)
)

type openOpts struct {
	bare bool
}

// OpenOption is an option which can be passed to the OpenRepo function to
// affect the Repo's behavior.
type OpenOption func(*openOpts)

// OpenBare returns an OpenOption which, if true is given, causes the OpenRepo
// function to expect to open a bare repo.
func OpenBare(bare bool) OpenOption {
	return func(o *openOpts) {
		o.bare = bare
	}
}

// Repo is an object which allows accessing and modifying the dehub repo.
type Repo struct {
	GitRepo *git.Repository

	// GitDirFS corresponds to the .git directory (or the entire repo directory
	// if it's a bare repo)
	GitDirFS billy.Filesystem
}

func extractGitDirFS(storer storage.Storer) (billy.Filesystem, error) {
	dotGitFSer, ok := storer.(interface{ Filesystem() billy.Filesystem })
	if !ok {
		return nil, fmt.Errorf("git storage object of type %T does not expose its underlying filesystem",
			storer)
	}
	return dotGitFSer.Filesystem(), nil
}

// OpenRepo opens the dehub repo in the given directory and returns the object
// for it.
//
// The given path is expected to have a git repo and .dehub folder already
// initialized.
func OpenRepo(path string, options ...OpenOption) (*Repo, error) {
	var opts openOpts
	for _, opt := range options {
		opt(&opts)
	}

	r := Repo{}
	var err error
	openOpts := &git.PlainOpenOptions{
		DetectDotGit: !opts.bare,
	}
	if r.GitRepo, err = git.PlainOpenWithOptions(path, openOpts); err != nil {
		return nil, fmt.Errorf("could not open git repo: %w", err)

	} else if r.GitDirFS, err = extractGitDirFS(r.GitRepo.Storer); err != nil {
		return nil, err
	}

	return &r, nil
}

type initOpts struct {
	bare   bool
	remote bool
}

// InitOption is an option which can be passed into the Init functions to affect
// their behavior.
type InitOption func(*initOpts)

// InitBare returns an InitOption which, if true is given, causes the Init
// function to initialize the repo without a worktree.
func InitBare(bare bool) InitOption {
	return func(o *initOpts) {
		o.bare = bare
	}
}

// InitRemote returns an InitOption which, if true is given, causes the Init
// function to initialize the repo with certain git configuration options set
// which make the repo able to be used as a remote repo.
func InitRemote(remote bool) InitOption {
	return func(o *initOpts) {
		o.remote = remote
	}
}

// InitRepo will initialize a new repository at the given path. If bare is true
// then the repository will not have a worktree.
func InitRepo(path string, options ...InitOption) (*Repo, error) {
	var opts initOpts
	for _, opt := range options {
		opt(&opts)
	}

	var repo Repo
	var err error
	if repo.GitRepo, err = git.PlainInit(path, opts.bare); err != nil {
		return nil, fmt.Errorf("initializing git repo: %w", err)

	} else if repo.GitDirFS, err = extractGitDirFS(repo.GitRepo.Storer); err != nil {
		return nil, err

	} else if err = repo.init(opts); err != nil {
		return nil, fmt.Errorf("initializing repo with dehub defaults: %w", err)
	}

	return &repo, nil
}

// InitMemRepo initializes an empty repository which only exists in memory.
func InitMemRepo(options ...InitOption) *Repo {
	var opts initOpts
	for _, opt := range options {
		opt(&opts)
	}

	fs := memfs.New()
	dotGitFS, err := fs.Chroot(git.GitDirName)
	if err != nil {
		panic(err)
	}
	storage := filesystem.NewStorage(dotGitFS, cache.NewObjectLRUDefault())

	var worktree billy.Filesystem
	if !opts.bare {
		worktree = fs
	}

	r, err := git.Init(storage, worktree)
	if err != nil {
		panic(err)
	}

	repo := &Repo{GitRepo: r, GitDirFS: dotGitFS}
	if err := repo.init(opts); err != nil {
		panic(err)
	}
	return repo
}

func (r *Repo) initRemotePreReceive(bare bool) error {
	if err := r.GitDirFS.MkdirAll("hooks", 0755); err != nil {
		return fmt.Errorf("creating hooks directory: %w", err)
	}

	preRcvFlags := os.O_WRONLY | os.O_CREATE | os.O_TRUNC
	preRcv, err := r.GitDirFS.OpenFile("hooks/pre-receive", preRcvFlags, 0755)
	if err != nil {
		return fmt.Errorf("opening hooks/pre-receive file: %w", err)
	}
	defer preRcv.Close()

	var preRcvBody string
	if bare {
		preRcvBody = "#!/bin/sh\nexec dehub hook -bare -pre-receive\n"
	} else {
		preRcvBody = "#!/bin/sh\nexec dehub hook -pre-receive\n"
	}

	if _, err := io.Copy(preRcv, bytes.NewBufferString(preRcvBody)); err != nil {
		return fmt.Errorf("writing to hooks/pre-receive: %w", err)
	}
	return nil
}

func (r *Repo) init(opts initOpts) error {
	headRef := plumbing.NewSymbolicReference(plumbing.HEAD, MainRefName)
	if err := r.GitRepo.Storer.SetReference(headRef); err != nil {
		return fmt.Errorf("setting HEAD reference to %q: %w", MainRefName, err)
	}

	if opts.remote {
		cfg, err := r.GitRepo.Config()
		if err != nil {
			return fmt.Errorf("opening git cfg: %w", err)
		}

		cfg.Raw = cfg.Raw.AddOption("http", config.NoSubsection, "receivepack", "true")
		if err := r.GitRepo.Storer.SetConfig(cfg); err != nil {
			return fmt.Errorf("storing modified git config: %w", err)
		}

		if err := r.initRemotePreReceive(opts.bare); err != nil {
			return fmt.Errorf("initializing pre-receive hook for remote-enabled repo: %w", err)
		}
	}

	return nil
}

func (r *Repo) billyFilesystem() (billy.Filesystem, error) {
	w, err := r.GitRepo.Worktree()
	if err != nil {
		return nil, fmt.Errorf("opening git worktree: %w", err)
	}
	return w.Filesystem, nil
}

var errTraverseRefNoMatch = errors.New("failed to find reference matching given predicate")

// TraverseReferenceChain resolves a chain of references, calling the given
// predicate on each one, and returning the first one for which the predicate
// returns true. This method will return an error if it reaches the end of the
// chain and the predicate still has not returned true.
//
// If a reference name is encountered which does not actually exist, then it is
// assumed to be a hash reference to the zero hash.
func (r *Repo) TraverseReferenceChain(refName plumbing.ReferenceName, pred func(*plumbing.Reference) bool) (*plumbing.Reference, error) {
	// TODO infinite loop checking
	for {
		ref, err := r.GitRepo.Storer.Reference(refName)
		if errors.Is(err, plumbing.ErrReferenceNotFound) {
			ref = plumbing.NewHashReference(refName, plumbing.ZeroHash)
		} else if err != nil {
			return nil, fmt.Errorf("resolving reference %q: %w", refName, err)
		}

		if pred(ref) {
			return ref, nil
		} else if ref.Type() != plumbing.SymbolicReference {
			return nil, errTraverseRefNoMatch
		}
		refName = ref.Target()
	}
}

// ErrNoBranchReference is returned from ReferenceToBranchName if no reference
// in the reference chain is for a branch.
var ErrNoBranchReference = errors.New("no branch reference found")

// ReferenceToBranchName traverses a chain of references looking for the first
// branch reference, and returns that name, or returns ErrNoBranchReference if
// no branch reference is part of the chain.
func (r *Repo) ReferenceToBranchName(refName plumbing.ReferenceName) (plumbing.ReferenceName, error) {
	// first check if the given refName is a branch, if so just return that.
	if refName.IsBranch() {
		return refName, nil
	}

	ref, err := r.TraverseReferenceChain(refName, func(ref *plumbing.Reference) bool {
		return ref.Target().IsBranch()
	})
	if errors.Is(err, errTraverseRefNoMatch) {
		return "", ErrNoBranchReference
	} else if err != nil {
		return "", fmt.Errorf("traversing reference chain: %w", err)
	}
	return ref.Target(), nil
}

// ReferenceToHash fully resolves a reference to a hash. If a reference cannot
// be resolved then plumbing.ZeroHash is returned.
func (r *Repo) ReferenceToHash(refName plumbing.ReferenceName) (plumbing.Hash, error) {
	ref, err := r.TraverseReferenceChain(refName, func(ref *plumbing.Reference) bool {
		return ref.Type() == plumbing.HashReference
	})
	if errors.Is(err, errTraverseRefNoMatch) {
		return plumbing.ZeroHash, errors.New("no hash in reference chain (is this even possible???)")
	} else if errors.Is(err, plumbing.ErrReferenceNotFound) {
		return plumbing.ZeroHash, nil
	} else if err != nil {
		return plumbing.ZeroHash, fmt.Errorf("traversing reference chain: %w", err)
	}
	return ref.Hash(), nil
}

// headFS returns an FS based on the HEAD commit, or if there is no HEAD commit
// (it's an empty repo) an FS based on the raw filesystem.
func (r *Repo) headFS() (fs.FS, error) {
	head, err := r.GetGitHead()
	if errors.Is(err, ErrHeadIsZero) {
		bfs, err := r.billyFilesystem()
		if err != nil {
			return nil, fmt.Errorf("getting underlying filesystem: %w", err)
		}
		return fs.FromBillyFilesystem(bfs), nil

	} else if err != nil {
		return nil, fmt.Errorf("could not get HEAD tree: %w", err)
	}

	return fs.FromTree(head.GitTree), nil
}

// GitCommit wraps a single git commit object, and also contains various fields
// which are parsed out of it. It is used as a convenience type, in place of
// having to manually retrieve and parse specific information out of commit
// objects.
type GitCommit struct {
	GitCommit *object.Commit

	// Fields based on that Commit, which can't be directly gleaned from it.
	GitTree   *object.Tree
	Commit    Commit
	Interface CommitInterface
}

// Root returns true if this commit is the root commit in its branch (i.e. it
// has no parents)
func (gc GitCommit) Root() bool {
	return gc.GitCommit.NumParents() == 0
}

// GetGitCommit retrieves the commit at the given hash, and all of its sub-data
// which can be pulled out of it.
func (r *Repo) GetGitCommit(h plumbing.Hash) (gc GitCommit, err error) {
	if gc.GitCommit, err = r.GitRepo.CommitObject(h); err != nil {
		return gc, fmt.Errorf("getting git commit object: %w", err)
	} else if gc.GitTree, err = r.GitRepo.TreeObject(gc.GitCommit.TreeHash); err != nil {
		return gc, fmt.Errorf("getting git tree object %q: %w",
			gc.GitCommit.TreeHash, err)
	} else if gc.Commit.UnmarshalText([]byte(gc.GitCommit.Message)); err != nil {
		return gc, fmt.Errorf("decoding commit message: %w", err)
	} else if gc.Interface, err = gc.Commit.Interface(); err != nil {
		return gc, fmt.Errorf("casting %+v to a CommitInterface: %w", gc.Commit, err)
	}
	return
}

// ErrHeadIsZero is used to indicate that HEAD resolves to the zero hash. An
// example of when this can happen is if the repo was just initialized and has
// no commits, or if an orphan branch is checked out.
var ErrHeadIsZero = errors.New("HEAD resolves to the zero hash")

// GetGitHead returns the GitCommit which is currently referenced by HEAD.
// This method may return ErrHeadIsZero if HEAD resolves to the zero hash.
func (r *Repo) GetGitHead() (GitCommit, error) {
	headHash, err := r.ReferenceToHash(plumbing.HEAD)
	if err != nil {
		return GitCommit{}, fmt.Errorf("resolving HEAD: %w", err)
	} else if headHash == plumbing.ZeroHash {
		return GitCommit{}, ErrHeadIsZero
	}

	gc, err := r.GetGitCommit(headHash)
	if err != nil {
		return GitCommit{}, fmt.Errorf("getting commit %q: %w", headHash, err)
	}
	return gc, nil
}

// GetGitCommitRange returns an ancestry of GitCommits, with the first being the
// commit immediately following the given starting hash, and the last being the
// given ending hash.
//
// If start is plumbing.ZeroHash then the root commit will be the starting one.
func (r *Repo) GetGitCommitRange(start, end plumbing.Hash) ([]GitCommit, error) {
	curr, err := r.GetGitCommit(end)
	if err != nil {
		return nil, fmt.Errorf("retrieving commit %q: %w", end, err)
	}

	var commits []GitCommit
	var found bool
	for {
		if found = start != plumbing.ZeroHash && curr.GitCommit.Hash == start; found {
			break
		}

		commits = append(commits, curr)
		numParents := curr.GitCommit.NumParents()
		if numParents == 0 {
			break
		} else if numParents > 1 {
			return nil, fmt.Errorf("commit %q has more than one parent: %+v",
				curr.GitCommit.Hash, curr.GitCommit.ParentHashes)
		}

		parentHash := curr.GitCommit.ParentHashes[0]
		parent, err := r.GetGitCommit(parentHash)
		if err != nil {
			return nil, fmt.Errorf("retrieving commit %q: %w", parentHash, err)
		}
		curr = parent
	}
	if !found && start != plumbing.ZeroHash {
		return nil, fmt.Errorf("unable to find commit %q as an ancestor of %q",
			start, end)
	}

	// reverse the commits to be in the expected order
	for l, r := 0, len(commits)-1; l < r; l, r = l+1, r-1 {
		commits[l], commits[r] = commits[r], commits[l]
	}
	return commits, nil
}

var (
	hashStrLen = len(plumbing.ZeroHash.String())
	errNotHex  = errors.New("not a valid hex string")
)

func (r *Repo) findCommitByShortHash(hashStr string) (plumbing.Hash, error) {
	paddedHashStr := hashStr
	if len(hashStr)%2 > 0 {
		paddedHashStr += "0"
	}

	if hashB, err := hex.DecodeString(paddedHashStr); err != nil {
		return plumbing.ZeroHash, errNotHex
	} else if len(hashStr) == hashStrLen {
		var hash plumbing.Hash
		copy(hash[:], hashB)
		return hash, nil
	} else if len(hashStr) < 2 {
		return plumbing.ZeroHash, errors.New("hash string must be 2 characters long or more")
	}

	for i := 2; i < hashStrLen; i++ {
		hashPrefix, hashTail := hashStr[:i], hashStr[i:]
		path := filepath.Join("objects", hashPrefix)
		fileInfos, err := r.GitDirFS.ReadDir(path)
		if err != nil {
			return plumbing.ZeroHash, fmt.Errorf("listing files in %q: %w", path, err)
		}

		var matchedHash plumbing.Hash
		for _, fileInfo := range fileInfos {
			objFileName := fileInfo.Name()
			if !strings.HasPrefix(objFileName, hashTail) {
				continue
			}

			objHash := plumbing.NewHash(hashPrefix + objFileName)
			obj, err := r.GitRepo.Storer.EncodedObject(plumbing.AnyObject, objHash)
			if err != nil {
				return plumbing.ZeroHash, fmt.Errorf("reading object %q off disk: %w", objHash, err)
			} else if obj.Type() != plumbing.CommitObject {
				continue

			} else if matchedHash == plumbing.ZeroHash {
				matchedHash = objHash
				continue
			}

			return plumbing.ZeroHash, fmt.Errorf("both %q and %q match", matchedHash, objHash)
		}

		if matchedHash != plumbing.ZeroHash {
			return matchedHash, nil
		}
	}

	return plumbing.ZeroHash, errors.New("failed to find a commit object with a matching prefix")
}

func (r *Repo) resolveRev(rev plumbing.Revision) (plumbing.Hash, error) {
	if rev == plumbing.Revision(plumbing.ZeroHash.String()) {
		return plumbing.ZeroHash, nil
	}

	{
		// pretend the revision is a short hash until proven otherwise
		shortHash := string(rev)
		hash, err := r.findCommitByShortHash(shortHash)
		if errors.Is(err, errNotHex) {
			// ok, continue
		} else if err != nil {
			return plumbing.ZeroHash, fmt.Errorf("resolving as short hash: %w", err)
		} else {
			// guess it _is_ a short hash, knew it!
			return hash, nil
		}
	}

	h, err := r.GitRepo.ResolveRevision(rev)
	if err != nil {
		return plumbing.ZeroHash, fmt.Errorf("resolving revision %q: %w", rev, err)
	}
	return *h, nil
}

// GetGitRevision resolves the revision and returns the GitCommit it references.
func (r *Repo) GetGitRevision(rev plumbing.Revision) (GitCommit, error) {
	hash, err := r.resolveRev(rev)
	if err != nil {
		return GitCommit{}, err
	}

	gc, err := r.GetGitCommit(hash)
	if err != nil {
		return GitCommit{}, fmt.Errorf("getting commit %q: %w", hash, err)
	}
	return gc, nil
}

// GetGitRevisionRange is like GetGitCommitRange, first resolving the given
// revisions into hashes before continuing with GetGitCommitRange's behavior.
func (r *Repo) GetGitRevisionRange(startRev, endRev plumbing.Revision) ([]GitCommit, error) {
	start, err := r.resolveRev(startRev)
	if err != nil {
		return nil, err
	}

	end, err := r.resolveRev(endRev)
	if err != nil {
		return nil, err
	}

	return r.GetGitCommitRange(start, end)
}