dirstate-v2: Reserve a few bytes of space for future extensions

See doc-comment

Differential Revision: https://phab.mercurial-scm.org/D11101

mercurial/dirstateutils/docket.py

@@ -1,75 +1,75 b''
 # dirstatedocket.py - docket file for dirstate-v2
 #
 # Copyright Mercurial Contributors
 #
 # This software may be used and distributed according to the terms of the
 # GNU General Public License version 2 or any later version.
 
 from __future__ import absolute_import
 
 import struct
 
 from ..revlogutils import docket as docket_mod
 
 
 V2_FORMAT_MARKER = b"dirstate-v2\n"
 
 # Must match the constant of the same name in
 # `rust/hg-core/src/dirstate_tree/on_disk.rs`
-TREE_METADATA_SIZE = 40
+TREE_METADATA_SIZE = 44
 
 # * 12 bytes: format marker
 # * 32 bytes: node ID of the working directory's first parent
 # * 32 bytes: node ID of the working directory's second parent
 # * 4 bytes: big-endian used size of the data file
 # * {TREE_METADATA_SIZE} bytes: tree metadata, parsed separately
 # * 1 byte: length of the data file's UUID
 # * variable: data file's UUID
 #
 # Node IDs are null-padded if shorter than 32 bytes.
 # A data file shorter than the specified used size is corrupted (truncated)
 HEADER = struct.Struct(
     ">{}s32s32sL{}sB".format(len(V2_FORMAT_MARKER), TREE_METADATA_SIZE)
 )
 
 
 class DirstateDocket(object):
     data_filename_pattern = b'dirstate.%s.d'
 
     def __init__(self, parents, data_size, tree_metadata, uuid):
         self.parents = parents
         self.data_size = data_size
         self.tree_metadata = tree_metadata
         self.uuid = uuid
 
     @classmethod
     def with_new_uuid(cls, parents, data_size, tree_metadata):
         return cls(parents, data_size, tree_metadata, docket_mod.make_uid())
 
     @classmethod
     def parse(cls, data, nodeconstants):
         if not data:
             parents = (nodeconstants.nullid, nodeconstants.nullid)
             return cls(parents, 0, b'', None)
         marker, p1, p2, data_size, meta, uuid_size = HEADER.unpack_from(data)
         if marker != V2_FORMAT_MARKER:
             raise ValueError("expected dirstate-v2 marker")
         uuid = data[HEADER.size : HEADER.size + uuid_size]
         p1 = p1[: nodeconstants.nodelen]
         p2 = p2[: nodeconstants.nodelen]
         return cls((p1, p2), data_size, meta, uuid)
 
     def serialize(self):
         p1, p2 = self.parents
         header = HEADER.pack(
             V2_FORMAT_MARKER,
             p1,
             p2,
             self.data_size,
             self.tree_metadata,
             len(self.uuid),
         )
         return header + self.uuid
 
     def data_filename(self):
         return self.data_filename_pattern % self.uuid

rust/hg-core/src/dirstate_tree/on_disk.rs

@@ -1,753 +1,760 b''
 //! The "version 2" disk representation of the dirstate
 //!
 //! # File format
 //!
 //! In dirstate-v2 format, the `.hg/dirstate` file is a "docket that starts
 //! with a fixed-sized header whose layout is defined by the `DocketHeader`
 //! struct, followed by the data file identifier.
 //!
 //! A separate `.hg/dirstate.{uuid}.d` file contains most of the data. That
 //! file may be longer than the size given in the docket, but not shorter. Only
 //! the start of the data file up to the given size is considered. The
 //! fixed-size "root" of the dirstate tree whose layout is defined by the
 //! `Root` struct is found at the end of that slice of data.
 //!
 //! Its `root_nodes` field contains the slice (offset and length) to
 //! the nodes representing the files and directories at the root of the
 //! repository. Each node is also fixed-size, defined by the `Node` struct.
 //! Nodes in turn contain slices to variable-size paths, and to their own child
 //! nodes (if any) for nested files and directories.
 
 use crate::dirstate_tree::dirstate_map::{self, DirstateMap, NodeRef};
 use crate::dirstate_tree::path_with_basename::WithBasename;
 use crate::errors::HgError;
 use crate::utils::hg_path::HgPath;
 use crate::DirstateEntry;
 use crate::DirstateError;
 use crate::DirstateParents;
 use crate::EntryState;
 use bytes_cast::unaligned::{I32Be, I64Be, U16Be, U32Be};
 use bytes_cast::BytesCast;
 use format_bytes::format_bytes;
 use std::borrow::Cow;
 use std::convert::{TryFrom, TryInto};
 use std::time::{Duration, SystemTime, UNIX_EPOCH};
 
 /// Added at the start of `.hg/dirstate` when the "v2" format is used.
 /// This a redundant sanity check more than an actual "magic number" since
 /// `.hg/requires` already governs which format should be used.
 pub const V2_FORMAT_MARKER: &[u8; 12] = b"dirstate-v2\n";
 
 /// Keep space for 256-bit hashes
 const STORED_NODE_ID_BYTES: usize = 32;
 
 /// … even though only 160 bits are used for now, with SHA-1
 const USED_NODE_ID_BYTES: usize = 20;
 
 pub(super) const IGNORE_PATTERNS_HASH_LEN: usize = 20;
 pub(super) type IgnorePatternsHash = [u8; IGNORE_PATTERNS_HASH_LEN];
 
 /// Must match the constant of the same name in
 /// `mercurial/dirstateutils/docket.py`
-const TREE_METADATA_SIZE: usize = 40;
+const TREE_METADATA_SIZE: usize = 44;
 
 /// Make sure that size-affecting changes are made knowingly
 #[allow(unused)]
 fn static_assert_size_of() {
-    let _ = std::mem::transmute::<DocketHeader, [u8; 121]>;
     let _ = std::mem::transmute::<TreeMetadata, [u8; TREE_METADATA_SIZE]>;
+    let _ = std::mem::transmute::<DocketHeader, [u8; TREE_METADATA_SIZE + 81]>;
     let _ = std::mem::transmute::<Node, [u8; 43]>;
 }
 
 // Must match `HEADER` in `mercurial/dirstateutils/docket.py`
 #[derive(BytesCast)]
 #[repr(C)]
 struct DocketHeader {
     marker: [u8; V2_FORMAT_MARKER.len()],
     parent_1: [u8; STORED_NODE_ID_BYTES],
     parent_2: [u8; STORED_NODE_ID_BYTES],
 
     /// Counted in bytes
     data_size: Size,
 
     metadata: TreeMetadata,
 
     uuid_size: u8,
 }
 
 pub struct Docket<'on_disk> {
     header: &'on_disk DocketHeader,
     uuid: &'on_disk [u8],
 }
 
 #[derive(BytesCast)]
 #[repr(C)]
 struct TreeMetadata {
     root_nodes: ChildNodes,
     nodes_with_entry_count: Size,
     nodes_with_copy_source_count: Size,
 
     /// How many bytes of this data file are not used anymore
     unreachable_bytes: Size,
 
+    /// Current version always sets these bytes to zero when creating or
+    /// updating a dirstate. Future versions could assign some bits to signal
+    /// for example "the version that last wrote/updated this dirstate did so
+    /// in such and such way that can be relied on by versions that know to."
+    unused: [u8; 4],
+
     /// If non-zero, a hash of ignore files that were used for some previous
     /// run of the `status` algorithm.
     ///
     /// We define:
     ///
     /// * "Root" ignore files are `.hgignore` at the root of the repository if
     ///   it exists, and files from `ui.ignore.*` config. This set of files is
     ///   then sorted by the string representation of their path.
     /// * The "expanded contents" of an ignore files is the byte string made
     ///   by concatenating its contents with the "expanded contents" of other
     ///   files included with `include:` or `subinclude:` files, in inclusion
     ///   order. This definition is recursive, as included files can
     ///   themselves include more files.
     ///
     /// This hash is defined as the SHA-1 of the concatenation (in sorted
     /// order) of the "expanded contents" of each "root" ignore file.
     /// (Note that computing this does not require actually concatenating byte
     /// strings into contiguous memory, instead SHA-1 hashing can be done
     /// incrementally.)
     ignore_patterns_hash: IgnorePatternsHash,
 }
 
 #[derive(BytesCast)]
 #[repr(C)]
 pub(super) struct Node {
     full_path: PathSlice,
 
     /// In bytes from `self.full_path.start`
     base_name_start: PathSize,
 
     copy_source: OptPathSlice,
     children: ChildNodes,
     pub(super) descendants_with_entry_count: Size,
     pub(super) tracked_descendants_count: Size,
 
     /// Depending on the value of `state`:
     ///
     /// * A null byte: `data` is not used.
     ///
     /// * A `n`, `a`, `r`, or `m` ASCII byte: `state` and `data` together
     ///   represent a dirstate entry like in the v1 format.
     ///
     /// * A `d` ASCII byte: the bytes of `data` should instead be interpreted
     ///   as the `Timestamp` for the mtime of a cached directory.
     ///
     ///   The presence of this state means that at some point, this path in
     ///   the working directory was observed:
     ///
     ///   - To be a directory
     ///   - With the modification time as given by `Timestamp`
     ///   - That timestamp was already strictly in the past when observed,
     ///     meaning that later changes cannot happen in the same clock tick
     ///     and must cause a different modification time (unless the system
     ///     clock jumps back and we get unlucky, which is not impossible but
     ///     but deemed unlikely enough).
     ///   - All direct children of this directory (as returned by
     ///     `std::fs::read_dir`) either have a corresponding dirstate node, or
     ///     are ignored by ignore patterns whose hash is in
     ///     `TreeMetadata::ignore_patterns_hash`.
     ///
     ///   This means that if `std::fs::symlink_metadata` later reports the
     ///   same modification time and ignored patterns haven’t changed, a run
     ///   of status that is not listing ignored   files can skip calling
     ///   `std::fs::read_dir` again for this directory,   iterate child
     ///   dirstate nodes instead.
     state: u8,
     data: Entry,
 }
 
 #[derive(BytesCast, Copy, Clone)]
 #[repr(C)]
 struct Entry {
     mode: I32Be,
     mtime: I32Be,
     size: I32Be,
 }
 
 /// Duration since the Unix epoch
 #[derive(BytesCast, Copy, Clone, PartialEq)]
 #[repr(C)]
 pub(super) struct Timestamp {
     seconds: I64Be,
 
     /// In `0 .. 1_000_000_000`.
     ///
     /// This timestamp is later or earlier than `(seconds, 0)` by this many
     /// nanoseconds, if `seconds` is non-negative or negative, respectively.
     nanoseconds: U32Be,
 }
 
 /// Counted in bytes from the start of the file
 ///
 /// NOTE: not supporting `.hg/dirstate` files larger than 4 GiB.
 type Offset = U32Be;
 
 /// Counted in number of items
 ///
 /// NOTE: we choose not to support counting more than 4 billion nodes anywhere.
 type Size = U32Be;
 
 /// Counted in bytes
 ///
 /// NOTE: we choose not to support file names/paths longer than 64 KiB.
 type PathSize = U16Be;
 
 /// A contiguous sequence of `len` times `Node`, representing the child nodes
 /// of either some other node or of the repository root.
 ///
 /// Always sorted by ascending `full_path`, to allow binary search.
 /// Since nodes with the same parent nodes also have the same parent path,
 /// only the `base_name`s need to be compared during binary search.
 #[derive(BytesCast, Copy, Clone)]
 #[repr(C)]
 struct ChildNodes {
     start: Offset,
     len: Size,
 }
 
 /// A `HgPath` of `len` bytes
 #[derive(BytesCast, Copy, Clone)]
 #[repr(C)]
 struct PathSlice {
     start: Offset,
     len: PathSize,
 }
 
 /// Either nothing if `start == 0`, or a `HgPath` of `len` bytes
 type OptPathSlice = PathSlice;
 
 /// Unexpected file format found in `.hg/dirstate` with the "v2" format.
 ///
 /// This should only happen if Mercurial is buggy or a repository is corrupted.
 #[derive(Debug)]
 pub struct DirstateV2ParseError;
 
 impl From<DirstateV2ParseError> for HgError {
     fn from(_: DirstateV2ParseError) -> Self {
         HgError::corrupted("dirstate-v2 parse error")
     }
 }
 
 impl From<DirstateV2ParseError> for crate::DirstateError {
     fn from(error: DirstateV2ParseError) -> Self {
         HgError::from(error).into()
     }
 }
 
 impl<'on_disk> Docket<'on_disk> {
     pub fn parents(&self) -> DirstateParents {
         use crate::Node;
         let p1 = Node::try_from(&self.header.parent_1[..USED_NODE_ID_BYTES])
             .unwrap()
             .clone();
         let p2 = Node::try_from(&self.header.parent_2[..USED_NODE_ID_BYTES])
             .unwrap()
             .clone();
         DirstateParents { p1, p2 }
     }
 
     pub fn tree_metadata(&self) -> &[u8] {
         self.header.metadata.as_bytes()
     }
 
     pub fn data_size(&self) -> usize {
         // This `unwrap` could only panic on a 16-bit CPU
         self.header.data_size.get().try_into().unwrap()
     }
 
     pub fn data_filename(&self) -> String {
         String::from_utf8(format_bytes!(b"dirstate.{}.d", self.uuid)).unwrap()
     }
 }
 
 pub fn read_docket(
     on_disk: &[u8],
 ) -> Result<Docket<'_>, DirstateV2ParseError> {
     let (header, uuid) =
         DocketHeader::from_bytes(on_disk).map_err(|_| DirstateV2ParseError)?;
     let uuid_size = header.uuid_size as usize;
     if header.marker == *V2_FORMAT_MARKER && uuid.len() == uuid_size {
         Ok(Docket { header, uuid })
     } else {
         Err(DirstateV2ParseError)
     }
 }
 
 pub(super) fn read<'on_disk>(
     on_disk: &'on_disk [u8],
     metadata: &[u8],
 ) -> Result<DirstateMap<'on_disk>, DirstateV2ParseError> {
     if on_disk.is_empty() {
         return Ok(DirstateMap::empty(on_disk));
     }
     let (meta, _) = TreeMetadata::from_bytes(metadata)
         .map_err(|_| DirstateV2ParseError)?;
     let dirstate_map = DirstateMap {
         on_disk,
         root: dirstate_map::ChildNodes::OnDisk(read_nodes(
             on_disk,
             meta.root_nodes,
         )?),
         nodes_with_entry_count: meta.nodes_with_entry_count.get(),
         nodes_with_copy_source_count: meta.nodes_with_copy_source_count.get(),
         ignore_patterns_hash: meta.ignore_patterns_hash,
         unreachable_bytes: meta.unreachable_bytes.get(),
     };
     Ok(dirstate_map)
 }
 
 impl Node {
     pub(super) fn full_path<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk HgPath, DirstateV2ParseError> {
         read_hg_path(on_disk, self.full_path)
     }
 
     pub(super) fn base_name_start<'on_disk>(
         &self,
     ) -> Result<usize, DirstateV2ParseError> {
         let start = self.base_name_start.get();
         if start < self.full_path.len.get() {
             let start = usize::try_from(start)
                 // u32 -> usize, could only panic on a 16-bit CPU
                 .expect("dirstate-v2 base_name_start out of bounds");
             Ok(start)
         } else {
             Err(DirstateV2ParseError)
         }
     }
 
     pub(super) fn base_name<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk HgPath, DirstateV2ParseError> {
         let full_path = self.full_path(on_disk)?;
         let base_name_start = self.base_name_start()?;
         Ok(HgPath::new(&full_path.as_bytes()[base_name_start..]))
     }
 
     pub(super) fn path<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<dirstate_map::NodeKey<'on_disk>, DirstateV2ParseError> {
         Ok(WithBasename::from_raw_parts(
             Cow::Borrowed(self.full_path(on_disk)?),
             self.base_name_start()?,
         ))
     }
 
     pub(super) fn has_copy_source<'on_disk>(&self) -> bool {
         self.copy_source.start.get() != 0
     }
 
     pub(super) fn copy_source<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<Option<&'on_disk HgPath>, DirstateV2ParseError> {
         Ok(if self.has_copy_source() {
             Some(read_hg_path(on_disk, self.copy_source)?)
         } else {
             None
         })
     }
 
     pub(super) fn node_data(
         &self,
     ) -> Result<dirstate_map::NodeData, DirstateV2ParseError> {
         let entry = |state| {
             dirstate_map::NodeData::Entry(self.entry_with_given_state(state))
         };
 
         match self.state {
             b'\0' => Ok(dirstate_map::NodeData::None),
             b'd' => Ok(dirstate_map::NodeData::CachedDirectory {
                 mtime: *self.data.as_timestamp(),
             }),
             b'n' => Ok(entry(EntryState::Normal)),
             b'a' => Ok(entry(EntryState::Added)),
             b'r' => Ok(entry(EntryState::Removed)),
             b'm' => Ok(entry(EntryState::Merged)),
             _ => Err(DirstateV2ParseError),
         }
     }
 
     pub(super) fn cached_directory_mtime(&self) -> Option<&Timestamp> {
         if self.state == b'd' {
             Some(self.data.as_timestamp())
         } else {
             None
         }
     }
 
     pub(super) fn state(
         &self,
     ) -> Result<Option<EntryState>, DirstateV2ParseError> {
         match self.state {
             b'\0' | b'd' => Ok(None),
             b'n' => Ok(Some(EntryState::Normal)),
             b'a' => Ok(Some(EntryState::Added)),
             b'r' => Ok(Some(EntryState::Removed)),
             b'm' => Ok(Some(EntryState::Merged)),
             _ => Err(DirstateV2ParseError),
         }
     }
 
     fn entry_with_given_state(&self, state: EntryState) -> DirstateEntry {
         DirstateEntry {
             state,
             mode: self.data.mode.get(),
             mtime: self.data.mtime.get(),
             size: self.data.size.get(),
         }
     }
 
     pub(super) fn entry(
         &self,
     ) -> Result<Option<DirstateEntry>, DirstateV2ParseError> {
         Ok(self
             .state()?
             .map(|state| self.entry_with_given_state(state)))
     }
 
     pub(super) fn children<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<&'on_disk [Node], DirstateV2ParseError> {
         read_nodes(on_disk, self.children)
     }
 
     pub(super) fn to_in_memory_node<'on_disk>(
         &self,
         on_disk: &'on_disk [u8],
     ) -> Result<dirstate_map::Node<'on_disk>, DirstateV2ParseError> {
         Ok(dirstate_map::Node {
             children: dirstate_map::ChildNodes::OnDisk(
                 self.children(on_disk)?,
             ),
             copy_source: self.copy_source(on_disk)?.map(Cow::Borrowed),
             data: self.node_data()?,
             descendants_with_entry_count: self
                 .descendants_with_entry_count
                 .get(),
             tracked_descendants_count: self.tracked_descendants_count.get(),
         })
     }
 }
 
 impl Entry {
     fn from_timestamp(timestamp: Timestamp) -> Self {
         // Safety: both types implement the `ByteCast` trait, so we could
         // safely use `as_bytes` and `from_bytes` to do this conversion. Using
         // `transmute` instead makes the compiler check that the two types
         // have the same size, which eliminates the error case of
         // `from_bytes`.
         unsafe { std::mem::transmute::<Timestamp, Entry>(timestamp) }
     }
 
     fn as_timestamp(&self) -> &Timestamp {
         // Safety: same as above in `from_timestamp`
         unsafe { &*(self as *const Entry as *const Timestamp) }
     }
 }
 
 impl Timestamp {
     pub fn seconds(&self) -> i64 {
         self.seconds.get()
     }
 }
 
 impl From<SystemTime> for Timestamp {
     fn from(system_time: SystemTime) -> Self {
         let (secs, nanos) = match system_time.duration_since(UNIX_EPOCH) {
             Ok(duration) => {
                 (duration.as_secs() as i64, duration.subsec_nanos())
             }
             Err(error) => {
                 let negative = error.duration();
                 (-(negative.as_secs() as i64), negative.subsec_nanos())
             }
         };
         Timestamp {
             seconds: secs.into(),
             nanoseconds: nanos.into(),
         }
     }
 }
 
 impl From<&'_ Timestamp> for SystemTime {
     fn from(timestamp: &'_ Timestamp) -> Self {
         let secs = timestamp.seconds.get();
         let nanos = timestamp.nanoseconds.get();
         if secs >= 0 {
             UNIX_EPOCH + Duration::new(secs as u64, nanos)
         } else {
             UNIX_EPOCH - Duration::new((-secs) as u64, nanos)
         }
     }
 }
 
 fn read_hg_path(
     on_disk: &[u8],
     slice: PathSlice,
 ) -> Result<&HgPath, DirstateV2ParseError> {
     read_slice(on_disk, slice.start, slice.len.get()).map(HgPath::new)
 }
 
 fn read_nodes(
     on_disk: &[u8],
     slice: ChildNodes,
 ) -> Result<&[Node], DirstateV2ParseError> {
     read_slice(on_disk, slice.start, slice.len.get())
 }
 
 fn read_slice<T, Len>(
     on_disk: &[u8],
     start: Offset,
     len: Len,
 ) -> Result<&[T], DirstateV2ParseError>
 where
     T: BytesCast,
     Len: TryInto<usize>,
 {
     // Either `usize::MAX` would result in "out of bounds" error since a single
     // `&[u8]` cannot occupy the entire addess space.
     let start = start.get().try_into().unwrap_or(std::usize::MAX);
     let len = len.try_into().unwrap_or(std::usize::MAX);
     on_disk
         .get(start..)
         .and_then(|bytes| T::slice_from_bytes(bytes, len).ok())
         .map(|(slice, _rest)| slice)
         .ok_or_else(|| DirstateV2ParseError)
 }
 
 pub(crate) fn for_each_tracked_path<'on_disk>(
     on_disk: &'on_disk [u8],
     metadata: &[u8],
     mut f: impl FnMut(&'on_disk HgPath),
 ) -> Result<(), DirstateV2ParseError> {
     let (meta, _) = TreeMetadata::from_bytes(metadata)
         .map_err(|_| DirstateV2ParseError)?;
     fn recur<'on_disk>(
         on_disk: &'on_disk [u8],
         nodes: ChildNodes,
         f: &mut impl FnMut(&'on_disk HgPath),
     ) -> Result<(), DirstateV2ParseError> {
         for node in read_nodes(on_disk, nodes)? {
             if let Some(state) = node.state()? {
                 if state.is_tracked() {
                     f(node.full_path(on_disk)?)
                 }
             }
             recur(on_disk, node.children, f)?
         }
         Ok(())
     }
     recur(on_disk, meta.root_nodes, &mut f)
 }
 
 /// Returns new data and metadata, together with whether that data should be
 /// appended to the existing data file whose content is at
 /// `dirstate_map.on_disk` (true), instead of written to a new data file
 /// (false).
 pub(super) fn write(
     dirstate_map: &mut DirstateMap,
     can_append: bool,
 ) -> Result<(Vec<u8>, Vec<u8>, bool), DirstateError> {
     let append = can_append && dirstate_map.write_should_append();
 
     // This ignores the space for paths, and for nodes without an entry.
     // TODO: better estimate? Skip the `Vec` and write to a file directly?
     let size_guess = std::mem::size_of::<Node>()
         * dirstate_map.nodes_with_entry_count as usize;
 
     let mut writer = Writer {
         dirstate_map,
         append,
         out: Vec::with_capacity(size_guess),
     };
 
     let root_nodes = writer.write_nodes(dirstate_map.root.as_ref())?;
 
     let meta = TreeMetadata {
         root_nodes,
         nodes_with_entry_count: dirstate_map.nodes_with_entry_count.into(),
         nodes_with_copy_source_count: dirstate_map
             .nodes_with_copy_source_count
             .into(),
         unreachable_bytes: dirstate_map.unreachable_bytes.into(),
+        unused: [0; 4],
         ignore_patterns_hash: dirstate_map.ignore_patterns_hash,
     };
     Ok((writer.out, meta.as_bytes().to_vec(), append))
 }
 
 struct Writer<'dmap, 'on_disk> {
     dirstate_map: &'dmap DirstateMap<'on_disk>,
     append: bool,
     out: Vec<u8>,
 }
 
 impl Writer<'_, '_> {
     fn write_nodes(
         &mut self,
         nodes: dirstate_map::ChildNodesRef,
     ) -> Result<ChildNodes, DirstateError> {
         // Reuse already-written nodes if possible
         if self.append {
             if let dirstate_map::ChildNodesRef::OnDisk(nodes_slice) = nodes {
                 let start = self.on_disk_offset_of(nodes_slice).expect(
                     "dirstate-v2 OnDisk nodes not found within on_disk",
                 );
                 let len = child_nodes_len_from_usize(nodes_slice.len());
                 return Ok(ChildNodes { start, len });
             }
         }
 
         // `dirstate_map::ChildNodes::InMemory` contains a `HashMap` which has
         // undefined iteration order. Sort to enable binary search in the
         // written file.
         let nodes = nodes.sorted();
         let nodes_len = nodes.len();
 
         // First accumulate serialized nodes in a `Vec`
         let mut on_disk_nodes = Vec::with_capacity(nodes_len);
         for node in nodes {
             let children =
                 self.write_nodes(node.children(self.dirstate_map.on_disk)?)?;
             let full_path = node.full_path(self.dirstate_map.on_disk)?;
             let full_path = self.write_path(full_path.as_bytes());
             let copy_source = if let Some(source) =
                 node.copy_source(self.dirstate_map.on_disk)?
             {
                 self.write_path(source.as_bytes())
             } else {
                 PathSlice {
                     start: 0.into(),
                     len: 0.into(),
                 }
             };
             on_disk_nodes.push(match node {
                 NodeRef::InMemory(path, node) => {
                     let (state, data) = match &node.data {
                         dirstate_map::NodeData::Entry(entry) => (
                             entry.state.into(),
                             Entry {
                                 mode: entry.mode.into(),
                                 mtime: entry.mtime.into(),
                                 size: entry.size.into(),
                             },
                         ),
                         dirstate_map::NodeData::CachedDirectory { mtime } => {
                             (b'd', Entry::from_timestamp(*mtime))
                         }
                         dirstate_map::NodeData::None => (
                             b'\0',
                             Entry {
                                 mode: 0.into(),
                                 mtime: 0.into(),
                                 size: 0.into(),
                             },
                         ),
                     };
                     Node {
                         children,
                         copy_source,
                         full_path,
                         base_name_start: u16::try_from(path.base_name_start())
                             // Could only panic for paths over 64 KiB
                             .expect("dirstate-v2 path length overflow")
                             .into(),
                         descendants_with_entry_count: node
                             .descendants_with_entry_count
                             .into(),
                         tracked_descendants_count: node
                             .tracked_descendants_count
                             .into(),
                         state,
                         data,
                     }
                 }
                 NodeRef::OnDisk(node) => Node {
                     children,
                     copy_source,
                     full_path,
                     ..*node
                 },
             })
         }
         // … so we can write them contiguously, after writing everything else
         // they refer to.
         let start = self.current_offset();
         let len = child_nodes_len_from_usize(nodes_len);
         self.out.extend(on_disk_nodes.as_bytes());
         Ok(ChildNodes { start, len })
     }
 
     /// If the given slice of items is within `on_disk`, returns its offset
     /// from the start of `on_disk`.
     fn on_disk_offset_of<T>(&self, slice: &[T]) -> Option<Offset>
     where
         T: BytesCast,
     {
         fn address_range(slice: &[u8]) -> std::ops::RangeInclusive<usize> {
             let start = slice.as_ptr() as usize;
             let end = start + slice.len();
             start..=end
         }
         let slice_addresses = address_range(slice.as_bytes());
         let on_disk_addresses = address_range(self.dirstate_map.on_disk);
         if on_disk_addresses.contains(slice_addresses.start())
             && on_disk_addresses.contains(slice_addresses.end())
         {
             let offset = slice_addresses.start() - on_disk_addresses.start();
             Some(offset_from_usize(offset))
         } else {
             None
         }
     }
 
     fn current_offset(&mut self) -> Offset {
         let mut offset = self.out.len();
         if self.append {
             offset += self.dirstate_map.on_disk.len()
         }
         offset_from_usize(offset)
     }
 
     fn write_path(&mut self, slice: &[u8]) -> PathSlice {
         let len = path_len_from_usize(slice.len());
         // Reuse an already-written path if possible
         if self.append {
             if let Some(start) = self.on_disk_offset_of(slice) {
                 return PathSlice { start, len };
             }
         }
         let start = self.current_offset();
         self.out.extend(slice.as_bytes());
         PathSlice { start, len }
     }
 }
 
 fn offset_from_usize(x: usize) -> Offset {
     u32::try_from(x)
         // Could only panic for a dirstate file larger than 4 GiB
         .expect("dirstate-v2 offset overflow")
         .into()
 }
 
 fn child_nodes_len_from_usize(x: usize) -> Size {
     u32::try_from(x)
         // Could only panic with over 4 billion nodes
         .expect("dirstate-v2 slice length overflow")
         .into()
 }
 
 fn path_len_from_usize(x: usize) -> PathSize {
     u16::try_from(x)
         // Could only panic for paths over 64 KiB
         .expect("dirstate-v2 path length overflow")
         .into()
 }