Status rewrites the manifest only when it helps, and asks keys on demand 992467d31d26
A snapshot rewrites the manifest only when it would teach the next one something: other flags, a file found equal to the base, a new blob, or a path added or gone. An edited file is read again next time whatever the manifest says, so a status of a few edits on the Linux tree no longer writes 96 000 lines (0.15 to 0.35 s). A stale manifest is always safe: a file whose stat changed is read. The walk asks a text file's graph key only when the manifest knows the file with the same stat and a key, rather than being handed every key of the base worked out in advance (0.09 s); the snapshot keeps the keys of the files it read. refs #41
- Author
- Andy McCall @andymccall <[email protected]>
- When
- · 2026-09-30 10:53 UTC
- Signature
- ✔ signed by Andy McCall
- Hash
992467d31d263dd85ae5804816f5d39ee7a0421d28a3c9bed735a17b6e77d7c1- Depends on
- 0bdc3f91a39f 75167672ebcd 8b276e0106d1 982864068a59 c3069537d7c5 cc7a00565161 d0990c2bff74 d0d52107b074 e7ddd552b619
- Depended on by
- 236f35c0baba cd4eb55e3b3d 3604179679eb f9466021286a 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
- 11 atoms, 0 tree operations, 2 files
Browse the tree at this change
modified crates/oxy-core/src/repo/mod.rs +15 −5
| … | 129 unchanged lines |
||
| 130 | 130 | pub is_head: bool, |
|
| 131 | 131 | } |
|
| 132 | 132 | ||
| 133 | + | /// A text file's base graph key, by path, as a walk asks for it. |
|
| 134 | + | pub(crate) type KeyOf<'a> = dyn Fn(&str) -> Option<String> + Sync + 'a; |
|
| 135 | + | ||
| 133 | 136 | /// A file as found on disk or in an imported tree. |
|
| 134 | 137 | pub(crate) struct FsEntry { |
|
| 135 | 138 | pub kind: NodeKind, |
|
| … | 1019 unchanged lines |
||
| 1155 | 1158 | &self, |
|
| 1156 | 1159 | cache: Option<&prompt::Manifest>, |
|
| 1157 | 1160 | tracked: &BTreeMap<String, (NodeKind, Option<ObjectRef>)>, |
|
| 1158 | - | keys: Option<&HashMap<String, String>>, |
|
| 1161 | + | keys: Option<&KeyOf>, |
|
| 1159 | 1162 | ) -> Result<BTreeMap<String, FsEntry>> { |
|
| 1160 | 1163 | let nested = std::sync::Arc::new(std::sync::Mutex::new(Vec::new())); |
|
| 1161 | 1164 | let wb = self.walker(&nested); |
|
| … | 83 unchanged lines |
||
| 1245 | 1248 | .filter(|(r, _, _)| is_attributes(r)) |
|
| 1246 | 1249 | .all(|(rel, _, meta)| { |
|
| 1247 | 1250 | stat_cached(rel, meta, cache) |
|
| 1248 | - | .is_some_and(|e| e.graph.is_some() && e.graph.as_ref() == k.get(rel)) |
|
| 1251 | + | .is_some_and(|e| e.graph.is_some() && e.graph == k(rel)) |
|
| 1249 | 1252 | }); |
|
| 1250 | 1253 | let none_removed = tracked |
|
| 1251 | 1254 | .keys() |
|
| … | 7 unchanged lines |
||
| 1259 | 1262 | let read: Vec<(String, FsEntry)> = files |
|
| 1260 | 1263 | .par_iter() |
|
| 1261 | 1264 | .map(|(rel, path, meta)| { |
|
| 1265 | + | // A key is worked out only for a file the manifest |
|
| 1266 | + | // knows with the same stat and a key of its own. |
|
| 1262 | 1267 | let key = keys |
|
| 1263 | 1268 | .filter(|_| !is_attributes(rel)) |
|
| 1264 | - | .and_then(|k| k.get(rel)) |
|
| 1265 | - | .map(String::as_str); |
|
| 1266 | - | Ok((rel.clone(), read_file_entry(rel, path, meta, cache, key)?)) |
|
| 1269 | + | .filter(|_| { |
|
| 1270 | + | stat_cached(rel, meta, cache).is_some_and(|e| e.graph.is_some()) |
|
| 1271 | + | }) |
|
| 1272 | + | .and_then(|k| k(rel)); |
|
| 1273 | + | Ok(( |
|
| 1274 | + | rel.clone(), |
|
| 1275 | + | read_file_entry(rel, path, meta, cache, key.as_deref())?, |
|
| 1276 | + | )) |
|
| 1267 | 1277 | }) |
|
| 1268 | 1278 | .collect::<Result<Vec<_>>>()?; |
|
| 1269 | 1279 | for (rel, mut e) in read { |
|
| … | 446 unchanged lines |
modified crates/oxy-core/src/repo/snapshot.rs +40 −18
| … | 291 unchanged lines |
||
| 292 | 292 | let _t = crate::timing::phase("snapshot: tree view"); |
|
| 293 | 293 | base.tree().view() |
|
| 294 | 294 | }; |
|
| 295 | - | let keys_phase = crate::timing::phase("snapshot: graph keys"); |
|
| 296 | - | let keys: HashMap<String, String> = view |
|
| 297 | - | .entries |
|
| 298 | - | .iter() |
|
| 299 | - | .filter(|(_, e)| e.kind == NodeKind::File && !e.mode.opaque) |
|
| 300 | - | .filter_map(|(p, e)| { |
|
| 301 | - | let k = base.graph_key(&e.node)?; |
|
| 302 | - | Some((p.clone(), crate::hash::ObjectRef(k).to_hex())) |
|
| 303 | - | }) |
|
| 304 | - | .collect(); |
|
| 305 | - | drop(keys_phase); |
|
| 306 | 295 | let manifest = { |
|
| 307 | 296 | let _t = crate::timing::phase("snapshot: read manifest"); |
|
| 308 | 297 | self.read_manifest()?.map(|(_, _, m)| m) |
|
| … | 14 unchanged lines |
||
| 323 | 312 | } |
|
| 324 | 313 | t |
|
| 325 | 314 | }; |
|
| 315 | + | // A text file's graph key, worked out when the walk asks for it. |
|
| 316 | + | let key_of = |p: &str| -> Option<String> { |
|
| 317 | + | let e = view.entries.get(p)?; |
|
| 318 | + | if e.kind != NodeKind::File || e.mode.opaque { |
|
| 319 | + | return None; |
|
| 320 | + | } |
|
| 321 | + | Some(crate::hash::ObjectRef(base.graph_key(&e.node)?).to_hex()) |
|
| 322 | + | }; |
|
| 326 | 323 | let mut fs_entries = { |
|
| 327 | 324 | let _t = crate::timing::phase("snapshot: read working copy"); |
|
| 328 | - | self.walk_working_copy_with(manifest.as_deref(), &tracked, Some(&keys))? |
|
| 325 | + | self.walk_working_copy_with(manifest.as_deref(), &tracked, Some(&key_of))? |
|
| 329 | 326 | }; |
|
| 327 | + | // The keys of the files read, before building graphs drops them: a |
|
| 328 | + | // file found equal to its base rendering keeps its key. |
|
| 329 | + | let keys: HashMap<String, String> = fs_entries |
|
| 330 | + | .iter() |
|
| 331 | + | .filter(|(_, e)| !e.unchanged && e.kind == NodeKind::File) |
|
| 332 | + | .filter_map(|(p, _)| Some((p.clone(), key_of(p)?))) |
|
| 333 | + | .collect(); |
|
| 330 | 334 | let base_snapshot = { |
|
| 331 | 335 | let _t = crate::timing::phase("snapshot: load and render base"); |
|
| 332 | 336 | if partial { |
|
| … | 81 unchanged lines |
||
| 414 | 418 | drop(record); |
|
| 415 | 419 | // The prompt compares against this without recording anything. |
|
| 416 | 420 | let _t = crate::timing::phase("snapshot: write manifest"); |
|
| 417 | - | self.write_manifest_entries( |
|
| 418 | - | base_snapshot.is_conflicted(), |
|
| 419 | - | !wc.tree_ops.is_empty() || !wc.edits.is_empty(), |
|
| 420 | - | &fs_entries, |
|
| 421 | - | &built.blobs, |
|
| 422 | - | )?; |
|
| 421 | + | let conflicted = base_snapshot.is_conflicted(); |
|
| 422 | + | let uncommitted = !wc.tree_ops.is_empty() || !wc.edits.is_empty(); |
|
| 423 | + | // Rewriting the manifest pays only when it would teach the next |
|
| 424 | + | // snapshot something: other flags, a file found equal to the base |
|
| 425 | + | // (it need not be read again), a new blob, a path added or gone. |
|
| 426 | + | // An edited file is read next time whatever it says. |
|
| 427 | + | let worth = match self.read_manifest()? { |
|
| 428 | + | None => true, |
|
| 429 | + | Some((c, u, m)) => { |
|
| 430 | + | c != conflicted |
|
| 431 | + | || u != uncommitted |
|
| 432 | + | || m.entries.len() != fs_entries.len() |
|
| 433 | + | || fs_entries.iter().any(|(p, e)| { |
|
| 434 | + | !e.unchanged && (e.graph_key.is_some() || !m.entries.contains_key(p)) |
|
| 435 | + | }) |
|
| 436 | + | || built |
|
| 437 | + | .blobs |
|
| 438 | + | .iter() |
|
| 439 | + | .any(|(p, b)| m.entries.get(p).and_then(|e| e.blob) != Some(*b)) |
|
| 440 | + | } |
|
| 441 | + | }; |
|
| 442 | + | if worth { |
|
| 443 | + | self.write_manifest_entries(conflicted, uncommitted, &fs_entries, &built.blobs)?; |
|
| 444 | + | } |
|
| 423 | 445 | Ok(SnapshotResult { |
|
| 424 | 446 | wc_hash: new_hash, |
|
| 425 | 447 | wc, |
|
| … | 1929 unchanged lines |