Check out against the manifest, without the base state 0554aaa7ddef

After the manifest check or a snapshot the manifest says, for every
file on disk, which graph it renders (its graph key), whether it is
executable, and its blob when opaque. A checkout therefore needs only
the target: a text file is left alone when its manifest key is the
target's, an opaque file when its blob is; removals come from the
manifest and the working copy's own changes, and a nested repository
is recognised on disk. What the new state lacks is counted by a walk
from the old heads that stops at the target. Files left as they were
keep their known stat in the new manifest instead of being statted
again. A one-file switch on the Linux tip takes 2.1 s and 455 MB (from
3.3 s and 705 MB).

refs #35

Author: Andy McCall @andymccall <[email protected]>
When: 1 month ago · 2026-09-29 22:48 UTC
Signature: ✔ signed by Andy McCall
Hash: 0554aaa7ddef03e68e06b8ff9dfe1a2787602e26acdff916f956250a2e142287
Depends on: 135212de5370 1d360c91a289 75167672ebcd 982864068a59 a4fad5b085c7 b7f102fcbb17 c5ce0367eddf cc7a00565161 d3cd7ef95da6
Depended on by: 1966f5db15ee
In bookmarks: agents-in-readme bare-hashes carried-not-removed carry-identical doctor-trust identical-sides keyless-tags main merge-remote-bookmark only-help-103 paths-for-resolutions-and-amend push-notes show-stat subrepo-pin-state switch-says
In tags: v0.1.0 v0.10.0 v0.11.0 v0.2.0 v0.3.0 v0.3.1 v0.4.0 v0.4.1 v0.4.2 v0.4.3 v0.4.4 v0.4.5 v0.4.6 v0.5.0 v0.5.1 v0.6.0 v0.7.0 v0.8.0 v0.9.0
Edits: 21 atoms, 0 tree operations, 4 files

Browse the tree at this change

hide whitespace changes

modified crates/oxy-core/src/repo/checkout.rs +46 −43

modified crates/oxy-core/src/repo/mod.rs +4 −0

modified crates/oxy-core/src/repo/prompt.rs +6 −3

modified docs/benchmarks.md +9 −0

		…	40 unchanged lines
41	41		let _wc = self.lock_working_copy()?;
42	42		let _all = crate::timing::phase("checkout");
43	43		// A working copy clean by the manifest needs no snapshot; otherwise a
44		-	// status snapshot, which neither reads nor renders files the
45		-	// manifest proves unchanged, and whose base is history only
46		-	// (oxygen#35).
47		-	let (base, changed_paths) = if self.clean_by_manifest()? {
48		-	let (_, wc) = self.working_copy()?;
49		-	(self.load_state_history(&wc.deps)?, Vec::new())
	44	+	// status snapshot records what changed, and rewrites the manifest
	45	+	// with the graph key of every file it proves unchanged. Either way
	46	+	// the manifest then says, for each file on disk, which graph it
	47	+	// renders, so the base state itself is not needed (oxygen#35).
	48	+	let changed_paths: Vec<String> = if self.clean_by_manifest()? {
	49	+	Vec::new()
50	50		} else {
51	51		let snap = self.snapshot_status(command)?;
52	52		self.refuse_dirty(&snap, discard)?;
53		-	let paths = snap.statuses.iter().map(\|s\| s.path.clone()).collect();
54		-	(snap.base, paths)
	53	+	snap.statuses.iter().map(\|s\| s.path.clone()).collect()
55	54		};
56		-	let views = crate::timing::phase("checkout: manifest and base view");
	55	+	let (_, wc_now) = self.working_copy()?;
57	56		let old_manifest = self.read_manifest()?.map(\|(_, _, m)\| m).unwrap_or_default();
58		-	let bview = base.tree().view();
59		-	drop(views);
60	57		let changed: HashSet<&str> = changed_paths.iter().map(String::as_str).collect();
61	58		// The target's history; its file graphs are built a batch at a time
62	59		// for the files that must be written, and dropped after.
		…	9 unchanged lines
72	69		.problems(tview.entries.keys().map(String::as_str)),
73	70		);
74	71		report.conflicted = tview.is_conflicted();
75		-	report.lacks = base.changes().filter(\|(h, _)\| !target.contains(h)).count();
	72	+	report.lacks = self.count_lacking(&wc_now.deps, &target)?;
76	73
77	74		// Remove what the base or the working copy has and the target does
78	75		// not. A nested repository is never deleted (spec §7.8).
79	76		let mut dirs_to_prune: BTreeSet<PathBuf> = BTreeSet::new();
80		-	let gone: BTreeSet<&str> = bview
	77	+	let gone: BTreeSet<&str> = old_manifest
81	78		.entries
82		-	.iter()
83		-	.filter(\|(_, f)\| f.kind != NodeKind::Dir)
84		-	.map(\|(p, _)\| p.as_str())
	79	+	.keys()
	80	+	.map(String::as_str)
85	81		.chain(changed.iter().copied())
86	82		.filter(\|p\| !tview.entries.contains_key(*p))
87	83		.collect();
88	84		for path in gone {
89		-	if bview
90		-	.entries
91		-	.get(path)
92		-	.is_some_and(\|f\| f.kind == NodeKind::Subrepo)
93		-	{
	85	+	let dir = self.root.join(path);
	86	+	if dir.join(OXY_DIR).exists() \|\| dir.join(".git").exists() {
94	87		report
95	88		.notes
96	89		.push(format!("{path}: nested repository left in place"));
		…	20 unchanged lines
117	110		if e.kind == NodeKind::Dir \|\| !self.in_scope(path) {
118	111		continue;
119	112		}
120		-	let untouched = !changed.contains(path.as_str())
121		-	&& bview
122		-	.entries
123		-	.get(path)
124		-	.is_some_and(\|b\| b.node == e.node && b.kind == e.kind && b.mode == e.mode);
	113	+	// The manifest entry of a file the working copy has not changed,
	114	+	// with the right executable bit.
	115	+	let exec_ok =
	116	+	\|m: &prompt::ManifestEntry\| !self.platform.exec_bits \|\| m.exec == e.mode.exec;
	117	+	let entry = old_manifest
	118	+	.entries
	119	+	.get(path)
	120	+	.filter(\|m\| !changed.contains(path.as_str()) && exec_ok(m));
125	121		if e.kind == NodeKind::File && !e.mode.opaque {
126		-	// The manifest proves the file on disk equal to the base's
127		-	// rendering (its key is the base graph's, and its stat was
128		-	// checked just now), and the graph is the same in both
129		-	// states: nothing to read, render or write.
130		-	let key = base.graph_key(&e.node);
131		-	let same_graph = untouched
132		-	&& !is_attributes(path)
133		-	&& key.is_some()
134		-	&& key == target.graph_key(&e.node);
	122	+	// The manifest proves the file on disk equal to the rendering
	123	+	// of a graph whose key is the target's: nothing to read,
	124	+	// render or write.
	125	+	let key = target.graph_key(&e.node).map(\|k\| ObjectRef(k).to_hex());
135	126		let proven = \|m: &&prompt::ManifestEntry\| {
136		-	same_graph
137		-	&& m.graph.as_deref() == key.map(\|k\| ObjectRef(k).to_hex()).as_deref()
	127	+	!is_attributes(path) && key.is_some() && m.graph == key
138	128		};
139		-	match old_manifest.entries.get(path).filter(proven) {
	129	+	match entry.filter(proven) {
140	130		Some(m) => {
141	131		entries.insert(
142	132		path.clone(),
		…	1 unchanged lines
144	134		content_hash: Some(m.hash.clone()),
145	135		graph_key: m.graph.clone(),
146	136		unchanged: true,
	137	+	stat: Some((m.size, m.mtime_ns)),
147	138		..FsEntry::new(NodeKind::File, e.mode.exec, Vec::new())
148	139		},
149	140		);
		…	25 unchanged lines
175	166		report.notes.extend(sync.note(path));
176	167		continue;
177	168		}
178		-	let same_blob = untouched && base.live_blobs(&e.node) == blobs;
179		-	if let Some(m) = old_manifest
180		-	.entries
181		-	.get(path)
182		-	.filter(\|_\| same_blob && blobs.len() == 1)
	169	+	if let Some(m) = entry
	170	+	.filter(\|m\| e.kind == NodeKind::File && blobs.len() == 1 && m.blob == Some(blob))
183	171		{
184	172		entries.insert(
185	173		path.clone(),
186	174		FsEntry {
187	175		content_hash: Some(m.hash.clone()),
188	176		blob: (e.kind == NodeKind::File).then_some(blob),
	177	+	stat: Some((m.size, m.mtime_ns)),
189	178		..FsEntry::new(e.kind, e.mode.exec, Vec::new())
190	179		},
191	180		);
		…	141 unchanged lines
333	322		),
334	323		);
335	324		Ok(report)
	325	+	}
	326	+
	327	+	/// How many changes of the state at `from` the state `to` lacks: a walk
	328	+	/// down from `from` that stops at what `to` holds.
	329	+	fn count_lacking(&self, from: &[ChangeHash], to: &State) -> Result<usize> {
	330	+	let mut seen: HashSet<ChangeHash> = HashSet::new();
	331	+	let mut stack: Vec<ChangeHash> = from.to_vec();
	332	+	while let Some(h) = stack.pop() {
	333	+	if to.contains(&h) \|\| !seen.insert(h) {
	334	+	continue;
	335	+	}
	336	+	stack.extend(self.store.get_change_lazy(&h)?.0.deps);
	337	+	}
	338	+	Ok(seen.len())
336	339		}
337	340
338	341		/// Refuse to leave uncommitted work behind unless `discard`.
		…	109 unchanged lines

		…	148 unchanged lines
149	149		pub unchanged: bool,
150	150		/// The base graph key this file equals, for the next manifest.
151	151		pub graph_key: Option<String>,
	152	+	/// Size and mtime already known to be current (a checkout leaving the
	153	+	/// file as it was), so the manifest need not stat it again.
	154	+	pub stat: Option<(u64, u128)>,
152	155		}
153	156
154	157		impl FsEntry {
		…	6 unchanged lines
161	164		content_hash: None,
162	165		unchanged: false,
163	166		graph_key: None,
	167	+	stat: None,
164	168		}
165	169		}
166	170		}
		…	1505 unchanged lines

		…	217 unchanged lines
218	218		let mut newest_ns: u128 = 0;
219	219		for (path, e) in entries {
220	220		let full = self.root.join(path);
221		-	let (size, mtime_ns) = match fs::symlink_metadata(&full) {
222		-	Ok(m) => (m.len(), mtime_ns(&m)),
223		-	Err(_) => (e.content.len() as u64, 0),
	221	+	let (size, mtime_ns) = match e.stat {
	222	+	Some(known) => known,
	223	+	None => match fs::symlink_metadata(&full) {
	224	+	Ok(m) => (m.len(), mtime_ns(&m)),
	225	+	Err(_) => (e.content.len() as u64, 0),
	226	+	},
224	227		};
225	228		let hash = match &e.content_hash {
226	229		Some(h) => h.clone(),
		…	323 unchanged lines

		…	339 unchanged lines
340	340		the target's history 0.5 s each, their tree views 0.3 s each, planning
341	341		0.2 s, and writing the manifest 0.6 s.
342	342
	343	+	A checkout now needs no base state at all: after the manifest check or a
	344	+	snapshot, the manifest says which graph each file on disk renders. A
	345	+	file is skipped when its manifest key is the target's, and a file left
	346	+	as it was keeps its stat in the new manifest instead of being statted
	347	+	again. What the new state lacks is found by a walk from the old heads
	348	+	that stops at the target. A one-file switch takes 2.1 to 2.2 s and 455
	349	+	MB: the manifest check 0.4 s, the target's history 0.5 s and tree view
	350	+	0.3 s, planning 0.16 s, and writing the manifest 0.4 s.
	351	+
343	352		Before that, most of a switch was loading history: the base's and the target's
344	353		history each decode the one huge first change (1.2 s and 364 MB each).
345	354		A clone holds the whole bundle in memory.
		…	1 unchanged lines

`/`	Search
`g` `c`	Code
`g` `h`	History
`g` `b`	Bookmarks
`g` `t`	Trust
`y`	Permalink to this state (files and trees)
`?`	This sheet

Check out against the manifest, without the base state 0554aaa7ddef

Keyboard