steal.v

1//! Stealing: bringing another machine's saves in, and combining when both2//! machines moved.3//!4//! The order is always the same, and it is what makes an interrupted steal5//! harmless: read everything from the remote first, then land it in one6//! transaction, then write the working tree. Divergence is resolved per file —7//! a clash is never merged line by line, it is chosen.89module sync1011import compress.zlib12import os13import src.apply14import src.events15import src.history16import src.remote17import src.snapshot18import src.storage1920pub enum Side {21	mine22	theirs23}2425// Clash is one file both machines changed differently. `stat` is the short26// diff stat the terminal prompt shows, so the choice is informed.27pub struct Clash {28pub mut:29	path   string30	mine   snapshot.FileEntry31	theirs snapshot.FileEntry32	base   snapshot.FileEntry33	stat   string34}3536pub struct StealOptions {37pub mut:38	mine   bool // resolve every clash as mine39	theirs bool // resolve every clash as theirs40	// Asked once per clash when neither flag is set. Nil means "nobody to ask".41	ask fn (clash Clash) Side = unsafe { nil }42}4344pub struct StealReport {45pub mut:46	repo            string47	forward         bool48	fetched_saves   int49	fetched_objects int50	fetched_events  int51	clashes         []string52	kept_mine       int53	replayed        int54	removed         int55	head            string56	flagged         []events.Flag // incoming saves that are flagged57	// Set when a combine took this machine's saves off the line: loading it58	// brings back the tree from before the steal, less anything stripped.59	previous_head string60}6162pub const dirty_tree = 'save or checkpoint first — steal needs a clean tree'63pub const no_saves_here = 'this folder has files but no saves — steal a repo into an empty folder with "koh steal <name>"'6465// steal brings the remote's new saves in. Inside a repo with its own work this66// may have to combine; a repo with nothing of its own simply fast-forwards.67pub fn steal(store &storage.Store, mut backend remote.Backend, opts StealOptions) !StealReport {68	name := repo_name(store, mut backend)!69	if snapshot.restore_pending(store) {70		return error(snapshot.restore_unfinished)71	}72	head_before := history.head(store)73	// Combining rewrites the working tree, so unsaved work would be lost.74	if dirty(store, head_before)! {75		if head_before == none {76			return error(no_saves_here)77		}78		return error(dirty_tree)79	}80	fetched := fetch_back_to_known(store, mut backend, name)!81	mut report := StealReport{82		repo:            name83		fetched_saves:   fetched.saves.len84		fetched_objects: fetched.repo_objects.len85		fetched_events:  fetched.events_out.len86		head:            fetched.head87	}88	mut local_only := []history.Save{}89	if head := head_before {90		// The remote builds on `base`. If that save is here but no longer on91		// the line, neither line contains the other; jumping to the remote92		// would silently leave this line behind.93		local_only = saves_after(store, head.id, fetched.base) or {94			return error('the remote builds on save ${fetched.base}, which is not on this line — load the save you want, then offer')95		}96	}97	if fetched.base == '' && local_only.len > 0 {98		return error('the remote keeps a different history — it shares no save with this repo')99	}100101	strips_before := events.strip_events(store)!.len102	mut clashes := []Clash{}103	mut side_of := map[string]Side{}104	if fetched.saves.len > 0 && local_only.len > 0 {105		clashes = combine_clashes(store, fetched)!106		report.clashes = clashes.map(it.path)107		// Asked before the write lock: a person thinking must not stall every108		// agent's save on this repo.109		side_of = choose_sides(clashes, opts)!110	}111112	store.begin() or { return err }113	// The questions ran without the lock. If a save or an edit landed114	// meanwhile, the answers were about a line that no longer exists.115	if moved := changed_since(store, head_before, strips_before) {116		if moved {117			store.rollback()118			return error('this repo changed while steal was asking — run koh steal again')119		}120	} else {121		store.rollback()122		return err123	}124	final := (write_and_combine(store, fetched, local_only, clashes, side_of, mut report) or {125		store.rollback()126		return err127	})128	files := history.files_of(store, final.id) or {129		store.rollback()130		return err131	}132	snapshot.check_links(store.root, files) or {133		store.rollback()134		return err135	}136	// The files follow the commit; the mark lands with it, in one step.137	if head_before == none || final.id != head_before or { history.Save{} }.id {138		snapshot.mark_restore(store) or {139			store.rollback()140			return err141		}142	}143	store.commit() or { return err }144	report.head = final.id145	if local_only.len > 0 && report.fetched_saves > 0 {146		report.previous_head = local_only.last().id147	}148	flags := events.latest_flags(store)!149	for record in fetched.saves {150		if flag := flags[record.id] {151			if flag.flagged {152				report.flagged << flag153			}154		}155	}156157	if head := head_before {158		if final.id == head.id {159			// The line did not move, so neither does the working tree.160			return report161		}162		mut wanted := map[string]bool{}163		for file in files {164			wanted[file.path] = true165		}166		mut remove := []string{}167		for file in history.files_of(store, head.id)! {168			if file.path !in wanted {169				remove << file.path170			}171		}172		result := snapshot.restore_files(store, store.root, snapshot.load(store.root), files,173			remove)!174		report.removed = result.removed175	} else {176		snapshot.restore_files(store, store.root, snapshot.load(store.root), files, [])!177	}178	history.refresh_log_md(store)179	return report180}181182// write_and_combine lands a fetch and, when the local line has saves the remote183// lacks, replays them on top of the remote head. It runs inside the caller's184// transaction and returns the new head.185fn write_and_combine(store &storage.Store, fetched Fetched, local_only []history.Save, clashes []Clash, side_of map[string]Side, mut report StealReport) !history.Save {186	write_fetched(store, fetched)!187	if fetched.saves.len == 0 {188		// The remote head is already on this line: level, or this machine is189		// ahead and the offer that follows a steal sends the rest.190		return history.head(store) or { return error('this repo has no head') }191	}192	if local_only.len == 0 {193		// Nothing of ours is missing from the remote: the line simply moves on.194		history.set_head(store, fetched.head)!195		report.forward = true196		return history.head(store) or { return error('the remote head did not land') }197	}198	remote_files := history.files_of(store, fetched.head)!199	mut mine_files := []snapshot.FileEntry{}200	if our_head := history.head(store) {201		mine_files = history.files_of(store, our_head.id)!202	}203204	// A file the remote changed is a clash, decided below, or one this machine205	// left as it was or changed the same way, so the remote's version stands.206	// Replaying this machine's steps on it could only conflict: an edit it207	// later undid would no longer apply.208	base_files := history.files_of(store, fetched.base)!209	settled := changed_paths(base_files, remote_files)210211	// Take the remote's line first, so the replayed saves chain onto it.212	history.set_head(store, fetched.head)!213	mut current := remote_files.clone()214	for save in local_only {215		parent_files := if save.parent != '' {216			history.files_of(store, save.parent)!217		} else {218			[]snapshot.FileEntry{}219		}220		target_files := history.files_of(store, save.id)!221		plan := apply.three_way(current, parent_files, target_files, [], settled)222		if apply.conflicts(plan) {223			return error('combine could not apply ${plan.conflict} — it is not a clash file')224		}225		if plan.applied.len == 0 {226			// Everything this save touched was a file the remote changed.227			continue228		}229		history.record_save(store, '', plan.files, history.SaveInput{230			message:       save.message231			created_at:    save.created_at232			machine:       save.machine233			agent:         save.agent234			session:       save.session235			is_checkpoint: save.is_checkpoint236			replayed_from: save.id237		})!238		current = plan.files.clone()239		report.replayed++240	}241242	mut mine_paths := []string{}243	for clash in clashes {244		if side_of[clash.path] == .mine {245			mine_paths << clash.path246		}247	}248	if mine_paths.len > 0 {249		// One last save puts my versions of the chosen files back.250		final_plan := apply.three_way(current, current, mine_files, mine_paths, [])251		if final_plan.applied.len > 0 {252			history.record_save(store, '', final_plan.files, history.SaveInput{253				message: 'combine: kept mine for ${mine_paths.len} ${if mine_paths.len == 1 {254					'file'255				} else {256					'files'257				}}'258			})!259			report.kept_mine = mine_paths.len260		}261	}262	return history.head(store) or { return error('the combined line has no head') }263}264265// find_clashes lists the files both sides changed, differently. Files changed266// identically on both sides are not clashes: there is nothing to decide.267pub fn find_clashes(base []snapshot.FileEntry, mine []snapshot.FileEntry, theirs []snapshot.FileEntry) []Clash {268	base_map := snapshot.index_files(base)269	mine_map := snapshot.index_files(mine)270	theirs_map := snapshot.index_files(theirs)271	mut paths := mine_map.keys()272	for path in theirs_map.keys() {273		if path !in mine_map {274			paths << path275		}276	}277	paths.sort()278	mut clashes := []Clash{}279	for path in paths {280		here := mine_map[path] or {281			snapshot.FileEntry{282				path: path283			}284		}285		there := theirs_map[path] or {286			snapshot.FileEntry{287				path: path288			}289		}290		was := base_map[path] or {291			snapshot.FileEntry{292				path: path293			}294		}295		if same(here, there) {296			continue297		}298		if same(here, was) || same(there, was) {299			continue300		}301		clashes << Clash{302			path:   path303			mine:   here304			theirs: there305			base:   was306		}307	}308	return clashes309}310311// changed_paths lists the paths whose entry differs between two trees.312fn changed_paths(before []snapshot.FileEntry, after []snapshot.FileEntry) []string {313	was := snapshot.index_files(before)314	now := snapshot.index_files(after)315	mut paths := []string{}316	for path, entry in now {317		if !same(entry, was[path] or { snapshot.FileEntry{} }) {318			paths << path319		}320	}321	for path, _ in was {322		if path !in now {323			paths << path324		}325	}326	return paths327}328329// combine_clashes finds the files both lines changed, with the diff stat each330// prompt shows. It reads the remote head from the fetch, not the store, so it331// can run before anything is written.332fn combine_clashes(store &storage.Store, fetched Fetched) ![]Clash {333	base_files := if fetched.base != '' {334		history.files_of(store, fetched.base)!335	} else {336		[]snapshot.FileEntry{}337	}338	head := history.head(store) or { return []Clash{} }339	// The transaction applies the incoming strips before it replays, so the340	// clashes are found on the trees as they will be then, or the answers341	// would be about files that are no longer there.342	strips := fetched.events_out.filter(it.kind == events.kind_strip)343	// Base and mine are here already, so every strip reaches them (created 0).344	remote_head := fetched.saves.last()345	mut clashes := find_clashes(after_strips(base_files, fetched.base, 0, strips), after_strips(history.files_of(store,346		head.id)!, head.id, 0, strips), after_strips(to_files(remote_head), remote_head.id,347		remote_head.created_at, strips))348	for mut clash in clashes {349		mine := fetched_bytes(store, fetched, clash.mine)!350		theirs := fetched_bytes(store, fetched, clash.theirs)!351		delta := snapshot.diff_bytes(clash.path, 'modified', mine, theirs)!352		clash.stat = if delta.binary { 'binary' } else { '+${delta.added} -${delta.removed}' }353	}354	return clashes355}356357// after_strips is a save's file list once the given strip events are applied,358// by the same rule write_fetched uses.359fn after_strips(files []snapshot.FileEntry, save_id string, created_at i64, strips []events.Event) []snapshot.FileEntry {360	return files.filter(fn [strips, save_id, created_at] (file snapshot.FileEntry) bool {361		return !strips.any(reaches(it, save_id, created_at) && covers(it.path, file.path))362	})363}364365// fetched_bytes reads one side of a clash: from the fetch when it has just366// arrived, else from the store. An absent side is empty.367fn fetched_bytes(store &storage.Store, fetched Fetched, entry snapshot.FileEntry) ![]u8 {368	if entry.hash == '' {369		return []u8{}370	}371	if blob := fetched.repo_objects[entry.hash] {372		return zlib.decompress(blob)!373	}374	return store.get_object(entry.hash)!375}376377fn choose_sides(clashes []Clash, opts StealOptions) !map[string]Side {378	mut sides := map[string]Side{}379	if clashes.len == 0 {380		return sides381	}382	for clash in clashes {383		if opts.mine {384			sides[clash.path] = .mine385			continue386		}387		if opts.theirs {388			sides[clash.path] = .theirs389			continue390		}391		if opts.ask != unsafe { nil } {392			sides[clash.path] = opts.ask(clash)393			continue394		}395		return error('${clashes.len} files changed on both machines — run on a terminal to choose, or pass --mine or --theirs')396	}397	return sides398}399400fn same(a snapshot.FileEntry, b snapshot.FileEntry) bool {401	if a.hash == '' && b.hash == '' {402		return true403	}404	return a.hash == b.hash && a.executable == b.executable405}406407fn changed_since(store &storage.Store, before ?history.Save, strips_before int) !bool {408	if events.strip_events(store)!.len != strips_before {409		return true410	}411	now := history.head(store)412	if now or { history.Save{} }.id != before or { history.Save{} }.id {413		return true414	}415	return dirty(store, now)!416}417418// dirty reports whether the working tree holds anything a steal would419// overwrite. With no saves yet, that is every file but the untouched ignore420// list `koh init` wrote.421fn dirty(store &storage.Store, head ?history.Save) !bool {422	files := snapshot.walk(store, store.root, snapshot.load(store.root))!423	if saved := head {424		return !snapshot.same_tree(history.files_of(store, saved.id)!, files)425	}426	seed := storage.hash_hex(snapshot.default_ignore.bytes())427	return files.any(it.path != snapshot.ignore_file || it.hash != seed)428}429430// fetch_repo pulls a whole repo — head, saves, objects and events — for431// `koh steal <name>`, which lands a copy in a new folder.432pub fn fetch_repo(mut backend remote.Backend, name string) !Fetched {433	remote_head := backend.get(head_key(name))!434	if !remote_head.found {435		return error('the remote has no repo called "${name}"')436	}437	save_ids := backend.list('${name}/saves')!438	object_keys := backend.list('${name}/objects')!439	event_keys := backend.list('${name}/events')!440	mut fetched := Fetched{441		head: remote_head.body.bytestr().trim_space()442	}443	for key in object_keys {444		blob := backend.get(key)!445		if blob.found {446			fetched.repo_objects[os.base(key)] = blob.body447		}448	}449	for key in save_ids {450		body := backend.get(key)!451		if !body.found {452			continue453		}454		record := decode_save(body.body)!455		fetched.saves << record456	}457	fetched.saves = parent_first(fetched.saves)!458	for key in event_keys {459		body := backend.get(key)!460		if body.found {461			fetched.events_out << decode_event(body.body)!462		}463	}464	return fetched465}466467// parent_first orders saves so every parent lands before its children, as the468// foreign key on `parent` demands. Time cannot do it: a replayed save keeps its469// original, older time but sits on a newer parent.470fn parent_first(records []SaveRecord) ![]SaveRecord {471	mut by_id := map[string]SaveRecord{}472	for record in records {473		by_id[record.id] = record474	}475	mut placed := map[string]bool{}476	mut ordered := []SaveRecord{}477	for record in records {478		mut chain := []SaveRecord{}479		mut cursor := record.id480		for cursor != '' && cursor !in placed {481			if chain.len > records.len {482				return error('the save records on the remote form a loop')483			}484			current := by_id[cursor] or { return error('the remote is missing save ${cursor}') }485			chain << current486			cursor = current.parent487		}488		chain.reverse_in_place()489		for current in chain {490			placed[current.id] = true491			ordered << current492		}493	}494	return ordered495}