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 exchange1011import compress.zlib12import os13import src.apply14import src.events15import src.fmt16import src.history17import src.remote18import src.snapshot19import src.storage2021pub enum Side {22	mine23	theirs24}2526// Clash is one file both machines changed differently. `stat` is the short27// diff stat the terminal prompt shows, so the choice is informed.28pub struct Clash {29pub mut:30	path         string31	mine         snapshot.FileEntry32	theirs       snapshot.FileEntry33	base         snapshot.FileEntry34	mine_stat    string35	theirs_stat  string36	mine_patch   string37	theirs_patch string38	mine_save    history.Save39	theirs_save  history.Save40}4142pub struct StealPreview {43pub:44	repo      string45	base      string46	local     []history.Save47	remote    []SaveRecord48	changed   int49	automatic int50	identical int51	clashes   []Clash52	yes       bool53	all_side  string54}5556pub struct StealOptions {57pub mut:58	mine   bool // resolve every clash as mine59	theirs bool // resolve every clash as theirs60	// Asked once per clash when neither flag is set. Nil means "nobody to ask".61	ask   fn (clash Clash) Side                      = unsafe { nil }62	guide fn (preview StealPreview) !map[string]Side = unsafe { nil }63	yes   bool64}6566pub struct StealReport {67pub mut:68	repo            string69	forward         bool70	fetched_saves   int71	fetched_objects int72	fetched_events  int73	clashes         []string74	kept_mine       int75	replayed        int76	removed         int77	head            string78	flagged         []events.Flag // incoming saves that are flagged79	// Set when a combine took this machine's saves off the line: loading it80	// brings back the tree from before the steal, less anything stripped.81	previous_head string82	kept_versions []string83}8485pub const dirty_tree = 'save or checkpoint first — steal needs a clean tree'86pub const no_saves_here = 'this folder has files but no saves — steal a repo into an empty folder with "koh steal <name>"'8788// steal brings the remote's new saves in. Inside a repo with its own work this89// may have to combine; a repo with nothing of its own simply fast-forwards.90pub fn steal(store &storage.Store, mut backend remote.Backend, opts StealOptions) !StealReport {91	name := repo_name(store, mut backend)!92	if snapshot.restore_pending(store) {93		return error(snapshot.restore_unfinished)94	}95	head_before := history.head(store)96	// Combining rewrites the working tree, so unsaved work would be lost.97	if dirty(store, head_before)! {98		if head_before == none {99			return error(no_saves_here)100		}101		return error(dirty_tree)102	}103	fetched := fetch_back_to_known(store, mut backend, name)!104	mut report := StealReport{105		repo:            name106		fetched_saves:   fetched.saves.len107		fetched_objects: fetched.repo_objects.len108		fetched_events:  fetched.events_out.len109		head:            fetched.head110	}111	mut local_only := []history.Save{}112	if head := head_before {113		// The remote builds on `base`. If that save is here but no longer on114		// the line, neither line contains the other; jumping to the remote115		// would silently leave this line behind.116		local_only = saves_after(store, head.id, fetched.base) or {117			return error('the remote builds on save ${fetched.base}, which is not on this line — load the save you want, then offer')118		}119	}120	if fetched.base == '' && local_only.len > 0 {121		return error('the remote keeps a different history — it shares no save with this repo')122	}123	if fetched.saves.len > 0 {124		store.begin()!125		write_fetched_records(store, fetched) or {126			store.rollback()127			return err128		}129		store.commit()!130	}131132	strips_before := events.strip_events(store)!.len133	mut clashes := []Clash{}134	mut side_of := map[string]Side{}135	if fetched.saves.len > 0 && local_only.len > 0 {136		clashes = combine_clashes(store, fetched, local_only)!137		report.clashes = clashes.map(it.path)138		if opts.guide != unsafe { nil } {139			side_of = opts.guide(combine_preview(store, name, fetched, local_only, clashes,140				opts.yes, if opts.mine {141					'mine'142				} else if opts.theirs { 'theirs' } else { '' })!)!143		} else {144			// Asked before the write lock: a person thinking must not stall every145			// agent's save on this repo.146			side_of = choose_sides(clashes, opts)!147		}148	}149	for clash in clashes {150		if side_of[clash.path] == .theirs {151			report.kept_versions << '${clash.path}:${local_only.last().id}'152		}153	}154155	store.begin() or { return err }156	// The questions ran without the lock. If a save or an edit landed157	// meanwhile, the answers were about a line that no longer exists.158	if moved := changed_since(store, head_before, strips_before) {159		if moved {160			store.rollback()161			return error('this repo changed while steal was asking — run koh steal again')162		}163	} else {164		store.rollback()165		return err166	}167	final := (write_and_combine(store, fetched, local_only, clashes, side_of, mut report) or {168		store.rollback()169		return err170	})171	files := history.files_of(store, final.id) or {172		store.rollback()173		return err174	}175	snapshot.check_links(store.root, files) or {176		store.rollback()177		return err178	}179	// The files follow the commit; the mark lands with it, in one step.180	// No head before means a fresh clone, whose whole tree is new.181	moved := if head := head_before { final.id != head.id } else { true }182	if moved {183		snapshot.mark_restore(store) or {184			store.rollback()185			return err186		}187	}188	store.commit() or { return err }189	report.head = final.id190	if local_only.len > 0 && report.fetched_saves > 0 {191		report.previous_head = local_only.last().id192	}193	flags := events.latest_flags(store)!194	for record in fetched.saves {195		if flag := flags[record.id] {196			if flag.flagged {197				report.flagged << flag198			}199		}200	}201202	if head := head_before {203		if final.id == head.id {204			// The line did not move, so neither does the working tree.205			return report206		}207		mut wanted := map[string]bool{}208		for file in files {209			wanted[file.path] = true210		}211		mut remove := []string{}212		for file in history.files_of(store, head.id)! {213			if file.path !in wanted {214				remove << file.path215			}216		}217		result := snapshot.restore_files(store, store.root, snapshot.load(store.root), files,218			remove)!219		report.removed = result.removed220	} else {221		snapshot.restore_files(store, store.root, snapshot.load(store.root), files, [])!222	}223	history.refresh_log_md(store)224	return report225}226227// write_and_combine lands a fetch and, when the local line has saves the remote228// lacks, replays them on top of the remote head. It runs inside the caller's229// transaction and returns the new head.230fn write_and_combine(store &storage.Store, fetched Fetched, local_only []history.Save, clashes []Clash, side_of map[string]Side, mut report StealReport) !history.Save {231	write_fetched(store, fetched)!232	if fetched.saves.len == 0 {233		// The remote head is already on this line: level, or this machine is234		// ahead and the offer that follows a steal sends the rest.235		return history.head(store) or { return error('this repo has no head') }236	}237	if local_only.len == 0 {238		// Nothing of ours is missing from the remote: the line simply moves on.239		history.set_head(store, fetched.head)!240		report.forward = true241		return history.head(store) or { return error('the remote head did not land') }242	}243	remote_files := history.files_of(store, fetched.head)!244	mut mine_files := []snapshot.FileEntry{}245	if our_head := history.head(store) {246		mine_files = history.files_of(store, our_head.id)!247	}248249	// A file the remote changed is a clash, decided below, or one this machine250	// left as it was or changed the same way, so the remote's version stands.251	// Replaying this machine's steps on it could only conflict: an edit it252	// later undid would no longer apply.253	base_files := history.files_of(store, fetched.base)!254	settled := changed_paths(base_files, remote_files)255256	// Take the remote's line first, so the replayed saves chain onto it.257	history.set_head(store, fetched.head)!258	mut current := remote_files.clone()259	for save in local_only {260		parent_files := if save.parent != '' {261			history.files_of(store, save.parent)!262		} else {263			[]snapshot.FileEntry{}264		}265		target_files := history.files_of(store, save.id)!266		plan := apply.three_way(current, parent_files, target_files, [], settled)267		if apply.conflicts(plan) {268			return error('combine could not apply ${plan.conflict} — it is not a clash file')269		}270		if plan.applied.len == 0 {271			// Everything this save touched was a file the remote changed.272			continue273		}274		history.record_save(store, '', plan.files, history.SaveInput{275			message:       save.message276			created_at:    save.created_at277			machine:       save.machine278			agent:         save.agent279			session:       save.session280			is_checkpoint: save.is_checkpoint281			replayed_from: save.id282		})!283		current = plan.files.clone()284		report.replayed++285	}286287	mut mine_paths := []string{}288	for clash in clashes {289		if side_of[clash.path] == .mine {290			mine_paths << clash.path291		}292	}293	if mine_paths.len > 0 {294		// One last save puts my versions of the chosen files back.295		final_plan := apply.three_way(current, current, mine_files, mine_paths, [])296		if final_plan.applied.len > 0 {297			history.record_save(store, '', final_plan.files, history.SaveInput{298				message: 'combine: kept mine for ${mine_paths.len} ${if mine_paths.len == 1 {299					'file'300				} else {301					'files'302				}}'303			})!304			report.kept_mine = mine_paths.len305		}306	}307	return history.head(store) or { return error('the combined line has no head') }308}309310// find_clashes lists the files both sides changed, differently. Files changed311// identically on both sides are not clashes: there is nothing to decide.312pub fn find_clashes(base []snapshot.FileEntry, mine []snapshot.FileEntry, theirs []snapshot.FileEntry) []Clash {313	base_map := snapshot.index_files(base)314	mine_map := snapshot.index_files(mine)315	theirs_map := snapshot.index_files(theirs)316	mut paths := mine_map.keys()317	for path in theirs_map.keys() {318		if path !in mine_map {319			paths << path320		}321	}322	paths.sort()323	mut clashes := []Clash{}324	for path in paths {325		here := mine_map[path] or {326			snapshot.FileEntry{327				path: path328			}329		}330		there := theirs_map[path] or {331			snapshot.FileEntry{332				path: path333			}334		}335		was := base_map[path] or {336			snapshot.FileEntry{337				path: path338			}339		}340		if same(here, there) {341			continue342		}343		if same(here, was) || same(there, was) {344			continue345		}346		clashes << Clash{347			path:   path348			mine:   here349			theirs: there350			base:   was351		}352	}353	return clashes354}355356// changed_paths lists the paths whose entry differs between two trees.357fn changed_paths(before []snapshot.FileEntry, after []snapshot.FileEntry) []string {358	was := snapshot.index_files(before)359	now := snapshot.index_files(after)360	mut paths := []string{}361	for path, entry in now {362		if !same(entry, was[path] or { snapshot.FileEntry{} }) {363			paths << path364		}365	}366	for path, _ in was {367		if path !in now {368			paths << path369		}370	}371	return paths372}373374// combine_clashes finds the files both lines changed, with the diff stat each375// prompt shows. It reads the remote head from the fetch, not the store, so it376// can run before anything is written.377fn combine_clashes(store &storage.Store, fetched Fetched, local_only []history.Save) ![]Clash {378	base_files := if fetched.base != '' {379		history.files_of(store, fetched.base)!380	} else {381		[]snapshot.FileEntry{}382	}383	head := history.head(store) or { return []Clash{} }384	// The transaction applies the incoming strips before it replays, so the385	// clashes are found on the trees as they will be then, or the answers386	// would be about files that are no longer there.387	strips := fetched.events_out.filter(it.kind == events.kind_strip)388	// Base and mine are here already, so every strip reaches them (created 0).389	remote_head := fetched.saves.last()390	mut clashes := find_clashes(after_strips(base_files, fetched.base, 0, strips), after_strips(history.files_of(store,391		head.id)!, head.id, 0, strips), after_strips(to_files(remote_head), remote_head.id,392		remote_head.created_at, strips))393	for mut clash in clashes {394		base := fetched_bytes(store, fetched, clash.base)!395		mine := fetched_bytes(store, fetched, clash.mine)!396		theirs := fetched_bytes(store, fetched, clash.theirs)!397		mine_delta := snapshot.diff_bytes(clash.path, 'modified', base, mine)!398		theirs_delta := snapshot.diff_bytes(clash.path, 'modified', base, theirs)!399		clash.mine_stat = delta_stat(mine_delta, clash.mine.size)400		clash.theirs_stat = delta_stat(theirs_delta, clash.theirs.size)401		clash.mine_patch = mine_delta.patch402		clash.theirs_patch = theirs_delta.patch403		clash.mine_save = last_touch(store, local_only, clash.path)!404		clash.theirs_save = last_touch(store, fetched.saves.map(to_save(it)), clash.path)!405	}406	return clashes407}408409fn delta_stat(delta snapshot.Delta, size i64) string {410	if delta.binary {411		return fmt.human_bytes(size)412	}413	return '+${delta.added} -${delta.removed}'414}415416fn last_touch(store &storage.Store, saves []history.Save, path string) !history.Save {417	for index := saves.len - 1; index >= 0; index-- {418		save := saves[index]419		current := snapshot.index_files(history.files_of(store, save.id)!)[path] or {420			snapshot.FileEntry{}421		}422		parent := if save.parent != '' {423			snapshot.index_files(history.files_of(store, save.parent)!)[path] or {424				snapshot.FileEntry{}425			}426		} else {427			snapshot.FileEntry{}428		}429		if !same(current, parent) {430			return save431		}432	}433	return saves.last()434}435436fn combine_preview(store &storage.Store, name string, fetched Fetched, local []history.Save, clashes []Clash, yes bool, all_side string) !StealPreview {437	base := history.files_of(store, fetched.base)!438	mine := history.files_of(store, local.last().id)!439	theirs := history.files_of(store, fetched.head)!440	mine_changed := changed_paths(base, mine)441	theirs_changed := changed_paths(base, theirs)442	mut changed := map[string]bool{}443	for path in mine_changed {444		changed[path] = true445	}446	for path in theirs_changed {447		changed[path] = true448	}449	mut identical := 0450	base_map := snapshot.index_files(base)451	mine_map := snapshot.index_files(mine)452	theirs_map := snapshot.index_files(theirs)453	for path, _ in changed {454		if !same(mine_map[path] or { snapshot.FileEntry{} }, base_map[path] or { snapshot.FileEntry{} })455			&& same(mine_map[path] or { snapshot.FileEntry{} }, theirs_map[path] or { snapshot.FileEntry{} }) {456			identical++457		}458	}459	return StealPreview{460		repo:      name461		base:      fetched.base462		local:     local463		remote:    fetched.saves464		changed:   changed.len465		automatic: changed.len - identical - clashes.len466		identical: identical467		clashes:   clashes468		yes:       yes469		all_side:  all_side470	}471}472473// after_strips is a save's file list once the given strip events are applied,474// by the same rule write_fetched uses.475fn after_strips(files []snapshot.FileEntry, save_id string, created_at i64, strips []events.Event) []snapshot.FileEntry {476	return files.filter(fn [strips, save_id, created_at] (file snapshot.FileEntry) bool {477		return !strips.any(reaches(it, save_id, created_at) && covers(it.path, file.path))478	})479}480481// fetched_bytes reads one side of a clash: from the fetch when it has just482// arrived, else from the store. An absent side is empty.483fn fetched_bytes(store &storage.Store, fetched Fetched, entry snapshot.FileEntry) ![]u8 {484	if entry.hash == '' {485		return []u8{}486	}487	if blob := fetched.repo_objects[entry.hash] {488		return zlib.decompress(blob)!489	}490	return store.get_object(entry.hash)!491}492493fn choose_sides(clashes []Clash, opts StealOptions) !map[string]Side {494	mut sides := map[string]Side{}495	if clashes.len == 0 {496		return sides497	}498	for clash in clashes {499		if opts.mine {500			sides[clash.path] = .mine501			continue502		}503		if opts.theirs {504			sides[clash.path] = .theirs505			continue506		}507		if opts.ask != unsafe { nil } {508			sides[clash.path] = opts.ask(clash)509			continue510		}511		paths := clashes.map(it.path).join(', ')512		return error('${paths} changed on both machines — run on a terminal to choose, or pass --mine or --theirs')513	}514	return sides515}516517fn same(a snapshot.FileEntry, b snapshot.FileEntry) bool {518	if a.hash == '' && b.hash == '' {519		return true520	}521	return a.hash == b.hash && a.executable == b.executable522}523524fn changed_since(store &storage.Store, before ?history.Save, strips_before int) !bool {525	if events.strip_events(store)!.len != strips_before {526		return true527	}528	now := history.head(store)529	if now or { history.Save{} }.id != before or { history.Save{} }.id {530		return true531	}532	return dirty(store, now)!533}534535// dirty reports whether the working tree holds anything a steal would536// overwrite. With no saves yet, that is every file but the untouched ignore537// list `koh init` wrote.538pub fn dirty(store &storage.Store, head ?history.Save) !bool {539	files := snapshot.walk(store, store.root, snapshot.load(store.root))!540	if saved := head {541		return !snapshot.same_tree(history.files_of(store, saved.id)!, files)542	}543	seed := storage.hash_hex(snapshot.default_ignore.bytes())544	return files.any(it.path != snapshot.ignore_file || it.hash != seed)545}546547// fetch_repo pulls a whole repo — head, saves, objects and events — for548// `koh steal <name>`, which lands a copy in a new folder.549pub fn fetch_repo(mut backend remote.Backend, name string) !Fetched {550	check_format(mut backend, name)!551	remote_head := backend.get(head_key(name))!552	if !remote_head.found {553		return error('the remote has no repo called "${name}"')554	}555	save_ids := backend.list('${name}/saves')!556	object_keys := backend.list('${name}/objects')!557	event_keys := backend.list('${name}/events')!558	mut fetched := Fetched{559		head: remote_head.body.bytestr().trim_space()560	}561	for key in object_keys {562		blob := backend.get(key)!563		if blob.found {564			fetched.repo_objects[os.base(key)] = blob.body565		}566	}567	for key in save_ids {568		body := backend.get(key)!569		if !body.found {570			continue571		}572		record := decode_save(body.body)!573		fetched.saves << record574	}575	fetched.saves = parent_first(fetched.saves)!576	for key in event_keys {577		body := backend.get(key)!578		if body.found {579			fetched.events_out << decode_event(body.body)!580		}581	}582	return fetched583}584585// parent_first orders saves so every parent lands before its children, as the586// foreign key on `parent` demands. Time cannot do it: a replayed save keeps its587// original, older time but sits on a newer parent.588fn parent_first(records []SaveRecord) ![]SaveRecord {589	mut by_id := map[string]SaveRecord{}590	for record in records {591		by_id[record.id] = record592	}593	mut placed := map[string]bool{}594	mut ordered := []SaveRecord{}595	for record in records {596		mut chain := []SaveRecord{}597		mut cursor := record.id598		for cursor != '' && cursor !in placed {599			if chain.len > records.len {600				return error('the save records on the remote form a loop')601			}602			current := by_id[cursor] or { return error('the remote is missing save ${cursor}') }603			chain << current604			cursor = current.parent605		}606		chain.reverse_in_place()607		for current in chain {608			placed[current.id] = true609			ordered << current610		}611	}612	return ordered613}