rhg: handle null changelog and manifest revisions

Differential Revision: https://phab.mercurial-scm.org/D11650

tests/test-empty-manifest-index.t

@@ -0,0 +1,23
+Create a repo such that the changelog entry refers to a null manifest node:
+
+  $ hg init a
+  $ cd a
+  $ hg log
+  $ touch x
+  $ hg add x
+  $ hg commit -m "init"
+  $ hg rm x
+  $ hg commit -q --amend
+
+  $ wc -c < .hg/store/00manifest.i
+  0
+
+Make sure that the manifest can be read (and is empty):
+
+  $ hg --config rhg.on-unsupported=abort files -r .
+  [1]
+
+Test a null changelog rev, too:
+
+  $ hg --config rhg.on-unsupported=abort files -r 0000000000000000000000000000000000000000
+  [1]

rust/hg-core/src/revlog/changelog.rs

@@ -1,67 +1,67
 use crate::errors::HgError;
 use crate::repo::Repo;
+use crate::revlog::node::NULL_NODE;
 use crate::revlog::revlog::{Revlog, RevlogError};
 use crate::revlog::Revision;
 use crate::revlog::{Node, NodePrefix};
 
 /// A specialized `Revlog` to work with `changelog` data format.
 pub struct Changelog {
     /// The generic `revlog` format.
     pub(crate) revlog: Revlog,
 }
 
 impl Changelog {
     /// Open the `changelog` of a repository given by its root.
     pub fn open(repo: &Repo) -> Result<Self, HgError> {
         let revlog = Revlog::open(repo, "00changelog.i", None)?;
         Ok(Self { revlog })
     }
 
     /// Return the `ChangelogEntry` for the given node ID.
     pub fn data_for_node(
         &self,
         node: NodePrefix,
     ) -> Result<ChangelogEntry, RevlogError> {
         let rev = self.revlog.rev_from_node(node)?;
         self.data_for_rev(rev)
     }
 
     /// Return the `ChangelogEntry` of the given revision number.
     pub fn data_for_rev(
         &self,
         rev: Revision,
     ) -> Result<ChangelogEntry, RevlogError> {
         let bytes = self.revlog.get_rev_data(rev)?;
         Ok(ChangelogEntry { bytes })
     }
 
     pub fn node_from_rev(&self, rev: Revision) -> Option<&Node> {
         self.revlog.node_from_rev(rev)
     }
 }
 
 /// `Changelog` entry which knows how to interpret the `changelog` data bytes.
 #[derive(Debug)]
 pub struct ChangelogEntry {
     /// The data bytes of the `changelog` entry.
     bytes: Vec<u8>,
 }
 
 impl ChangelogEntry {
     /// Return an iterator over the lines of the entry.
     pub fn lines(&self) -> impl Iterator<Item = &[u8]> {
         self.bytes
             .split(|b| b == &b'\n')
             .filter(|line| !line.is_empty())
     }
 
     /// Return the node id of the `manifest` referenced by this `changelog`
     /// entry.
     pub fn manifest_node(&self) -> Result<Node, HgError> {
-        Node::from_hex_for_repo(
-            self.lines()
-                .next()
-                .ok_or_else(|| HgError::corrupted("empty changelog entry"))?,
-        )
+        match self.lines().next() {
+            None => Ok(NULL_NODE),
+            Some(x) => Node::from_hex_for_repo(x),
     }
 }
+}

rust/hg-core/src/revlog/index.rs

@@ -1,403 +1,406
 use std::convert::TryInto;
 use std::ops::Deref;
 
 use byteorder::{BigEndian, ByteOrder};
 
 use crate::errors::HgError;
 use crate::revlog::node::Node;
 use crate::revlog::{Revision, NULL_REVISION};
 
 pub const INDEX_ENTRY_SIZE: usize = 64;
 
 /// A Revlog index
 pub struct Index {
     bytes: Box<dyn Deref<Target = [u8]> + Send>,
     /// Offsets of starts of index blocks.
     /// Only needed when the index is interleaved with data.
     offsets: Option<Vec<usize>>,
 }
 
 impl Index {
     /// Create an index from bytes.
     /// Calculate the start of each entry when is_inline is true.
     pub fn new(
         bytes: Box<dyn Deref<Target = [u8]> + Send>,
     ) -> Result<Self, HgError> {
         if is_inline(&bytes) {
             let mut offset: usize = 0;
             let mut offsets = Vec::new();
 
             while offset + INDEX_ENTRY_SIZE <= bytes.len() {
                 offsets.push(offset);
                 let end = offset + INDEX_ENTRY_SIZE;
                 let entry = IndexEntry {
                     bytes: &bytes[offset..end],
                     offset_override: None,
                 };
 
                 offset += INDEX_ENTRY_SIZE + entry.compressed_len();
             }
 
             if offset == bytes.len() {
                 Ok(Self {
                     bytes,
                     offsets: Some(offsets),
                 })
             } else {
                 Err(HgError::corrupted("unexpected inline revlog length")
                     .into())
             }
         } else {
             Ok(Self {
                 bytes,
                 offsets: None,
             })
         }
     }
 
     /// Value of the inline flag.
     pub fn is_inline(&self) -> bool {
         is_inline(&self.bytes)
     }
 
     /// Return a slice of bytes if `revlog` is inline. Panic if not.
     pub fn data(&self, start: usize, end: usize) -> &[u8] {
         if !self.is_inline() {
             panic!("tried to access data in the index of a revlog that is not inline");
         }
         &self.bytes[start..end]
     }
 
     /// Return number of entries of the revlog index.
     pub fn len(&self) -> usize {
         if let Some(offsets) = &self.offsets {
             offsets.len()
         } else {
             self.bytes.len() / INDEX_ENTRY_SIZE
         }
     }
 
     /// Returns `true` if the `Index` has zero `entries`.
     pub fn is_empty(&self) -> bool {
         self.len() == 0
     }
 
     /// Return the index entry corresponding to the given revision if it
     /// exists.
     pub fn get_entry(&self, rev: Revision) -> Option<IndexEntry> {
         if rev == NULL_REVISION {
             return None;
         }
         if let Some(offsets) = &self.offsets {
             self.get_entry_inline(rev, offsets)
         } else {
             self.get_entry_separated(rev)
         }
     }
 
     fn get_entry_inline(
         &self,
         rev: Revision,
         offsets: &[usize],
     ) -> Option<IndexEntry> {
         let start = *offsets.get(rev as usize)?;
         let end = start.checked_add(INDEX_ENTRY_SIZE)?;
         let bytes = &self.bytes[start..end];
 
         // See IndexEntry for an explanation of this override.
         let offset_override = Some(end);
 
         Some(IndexEntry {
             bytes,
             offset_override,
         })
     }
 
     fn get_entry_separated(&self, rev: Revision) -> Option<IndexEntry> {
         let max_rev = self.bytes.len() / INDEX_ENTRY_SIZE;
         if rev as usize >= max_rev {
             return None;
         }
         let start = rev as usize * INDEX_ENTRY_SIZE;
         let end = start + INDEX_ENTRY_SIZE;
         let bytes = &self.bytes[start..end];
 
         // Override the offset of the first revision as its bytes are used
         // for the index's metadata (saving space because it is always 0)
         let offset_override = if rev == 0 { Some(0) } else { None };
 
         Some(IndexEntry {
             bytes,
             offset_override,
         })
     }
 }
 
 impl super::RevlogIndex for Index {
     fn len(&self) -> usize {
         self.len()
     }
 
     fn node(&self, rev: Revision) -> Option<&Node> {
         self.get_entry(rev).map(|entry| entry.hash())
     }
 }
 
 #[derive(Debug)]
 pub struct IndexEntry<'a> {
     bytes: &'a [u8],
     /// Allows to override the offset value of the entry.
     ///
     /// For interleaved index and data, the offset stored in the index
     /// corresponds to the separated data offset.
     /// It has to be overridden with the actual offset in the interleaved
     /// index which is just after the index block.
     ///
     /// For separated index and data, the offset stored in the first index
     /// entry is mixed with the index headers.
     /// It has to be overridden with 0.
     offset_override: Option<usize>,
 }
 
 impl<'a> IndexEntry<'a> {
     /// Return the offset of the data.
     pub fn offset(&self) -> usize {
         if let Some(offset_override) = self.offset_override {
             offset_override
         } else {
             let mut bytes = [0; 8];
             bytes[2..8].copy_from_slice(&self.bytes[0..=5]);
             BigEndian::read_u64(&bytes[..]) as usize
         }
     }
 
     /// Return the compressed length of the data.
     pub fn compressed_len(&self) -> usize {
         BigEndian::read_u32(&self.bytes[8..=11]) as usize
     }
 
     /// Return the uncompressed length of the data.
     pub fn uncompressed_len(&self) -> usize {
         BigEndian::read_u32(&self.bytes[12..=15]) as usize
     }
 
     /// Return the revision upon which the data has been derived.
     pub fn base_revision(&self) -> Revision {
         // TODO Maybe return an Option when base_revision == rev?
         //      Requires to add rev to IndexEntry
 
         BigEndian::read_i32(&self.bytes[16..])
     }
 
     pub fn p1(&self) -> Revision {
         BigEndian::read_i32(&self.bytes[24..])
     }
 
     pub fn p2(&self) -> Revision {
         BigEndian::read_i32(&self.bytes[28..])
     }
 
     /// Return the hash of revision's full text.
     ///
     /// Currently, SHA-1 is used and only the first 20 bytes of this field
     /// are used.
     pub fn hash(&self) -> &'a Node {
         (&self.bytes[32..52]).try_into().unwrap()
     }
 }
 
 /// Value of the inline flag.
 pub fn is_inline(index_bytes: &[u8]) -> bool {
+    if index_bytes.len() < 4 {
+        return true;
+    }
     match &index_bytes[0..=1] {
         [0, 0] | [0, 2] => false,
         _ => true,
     }
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
 
     #[cfg(test)]
     #[derive(Debug, Copy, Clone)]
     pub struct IndexEntryBuilder {
         is_first: bool,
         is_inline: bool,
         is_general_delta: bool,
         version: u16,
         offset: usize,
         compressed_len: usize,
         uncompressed_len: usize,
         base_revision: Revision,
     }
 
     #[cfg(test)]
     impl IndexEntryBuilder {
         pub fn new() -> Self {
             Self {
                 is_first: false,
                 is_inline: false,
                 is_general_delta: true,
                 version: 2,
                 offset: 0,
                 compressed_len: 0,
                 uncompressed_len: 0,
                 base_revision: 0,
             }
         }
 
         pub fn is_first(&mut self, value: bool) -> &mut Self {
             self.is_first = value;
             self
         }
 
         pub fn with_inline(&mut self, value: bool) -> &mut Self {
             self.is_inline = value;
             self
         }
 
         pub fn with_general_delta(&mut self, value: bool) -> &mut Self {
             self.is_general_delta = value;
             self
         }
 
         pub fn with_version(&mut self, value: u16) -> &mut Self {
             self.version = value;
             self
         }
 
         pub fn with_offset(&mut self, value: usize) -> &mut Self {
             self.offset = value;
             self
         }
 
         pub fn with_compressed_len(&mut self, value: usize) -> &mut Self {
             self.compressed_len = value;
             self
         }
 
         pub fn with_uncompressed_len(&mut self, value: usize) -> &mut Self {
             self.uncompressed_len = value;
             self
         }
 
         pub fn with_base_revision(&mut self, value: Revision) -> &mut Self {
             self.base_revision = value;
             self
         }
 
         pub fn build(&self) -> Vec<u8> {
             let mut bytes = Vec::with_capacity(INDEX_ENTRY_SIZE);
             if self.is_first {
                 bytes.extend(&match (self.is_general_delta, self.is_inline) {
                     (false, false) => [0u8, 0],
                     (false, true) => [0u8, 1],
                     (true, false) => [0u8, 2],
                     (true, true) => [0u8, 3],
                 });
                 bytes.extend(&self.version.to_be_bytes());
                 // Remaining offset bytes.
                 bytes.extend(&[0u8; 2]);
             } else {
                 // Offset stored on 48 bits (6 bytes)
                 bytes.extend(&(self.offset as u64).to_be_bytes()[2..]);
             }
             bytes.extend(&[0u8; 2]); // Revision flags.
             bytes.extend(&(self.compressed_len as u32).to_be_bytes());
             bytes.extend(&(self.uncompressed_len as u32).to_be_bytes());
             bytes.extend(&self.base_revision.to_be_bytes());
             bytes
         }
     }
 
     #[test]
     fn is_not_inline_when_no_inline_flag_test() {
         let bytes = IndexEntryBuilder::new()
             .is_first(true)
             .with_general_delta(false)
             .with_inline(false)
             .build();
 
         assert_eq!(is_inline(&bytes), false)
     }
 
     #[test]
     fn is_inline_when_inline_flag_test() {
         let bytes = IndexEntryBuilder::new()
             .is_first(true)
             .with_general_delta(false)
             .with_inline(true)
             .build();
 
         assert_eq!(is_inline(&bytes), true)
     }
 
     #[test]
     fn is_inline_when_inline_and_generaldelta_flags_test() {
         let bytes = IndexEntryBuilder::new()
             .is_first(true)
             .with_general_delta(true)
             .with_inline(true)
             .build();
 
         assert_eq!(is_inline(&bytes), true)
     }
 
     #[test]
     fn test_offset() {
         let bytes = IndexEntryBuilder::new().with_offset(1).build();
         let entry = IndexEntry {
             bytes: &bytes,
             offset_override: None,
         };
 
         assert_eq!(entry.offset(), 1)
     }
 
     #[test]
     fn test_with_overridden_offset() {
         let bytes = IndexEntryBuilder::new().with_offset(1).build();
         let entry = IndexEntry {
             bytes: &bytes,
             offset_override: Some(2),
         };
 
         assert_eq!(entry.offset(), 2)
     }
 
     #[test]
     fn test_compressed_len() {
         let bytes = IndexEntryBuilder::new().with_compressed_len(1).build();
         let entry = IndexEntry {
             bytes: &bytes,
             offset_override: None,
         };
 
         assert_eq!(entry.compressed_len(), 1)
     }
 
     #[test]
     fn test_uncompressed_len() {
         let bytes = IndexEntryBuilder::new().with_uncompressed_len(1).build();
         let entry = IndexEntry {
             bytes: &bytes,
             offset_override: None,
         };
 
         assert_eq!(entry.uncompressed_len(), 1)
     }
 
     #[test]
     fn test_base_revision() {
         let bytes = IndexEntryBuilder::new().with_base_revision(1).build();
         let entry = IndexEntry {
             bytes: &bytes,
             offset_override: None,
         };
 
         assert_eq!(entry.base_revision(), 1)
     }
 }
 
 #[cfg(test)]
 pub use tests::IndexEntryBuilder;

rust/hg-core/src/revlog/revlog.rs

@@ -1,412 +1,423
 use std::borrow::Cow;
 use std::io::Read;
 use std::ops::Deref;
 use std::path::Path;
 
 use byteorder::{BigEndian, ByteOrder};
 use flate2::read::ZlibDecoder;
 use micro_timer::timed;
 use sha1::{Digest, Sha1};
 use zstd;
 
 use super::index::Index;
 use super::node::{NodePrefix, NODE_BYTES_LENGTH, NULL_NODE};
 use super::nodemap;
 use super::nodemap::{NodeMap, NodeMapError};
 use super::nodemap_docket::NodeMapDocket;
 use super::patch;
 use crate::errors::HgError;
 use crate::repo::Repo;
 use crate::revlog::Revision;
 use crate::{Node, NULL_REVISION};
 
 #[derive(derive_more::From)]
 pub enum RevlogError {
     InvalidRevision,
     /// Working directory is not supported
     WDirUnsupported,
     /// Found more than one entry whose ID match the requested prefix
     AmbiguousPrefix,
     #[from]
     Other(HgError),
 }
 
 impl From<NodeMapError> for RevlogError {
     fn from(error: NodeMapError) -> Self {
         match error {
             NodeMapError::MultipleResults => RevlogError::AmbiguousPrefix,
             NodeMapError::RevisionNotInIndex(_) => RevlogError::corrupted(),
         }
     }
 }
 
 impl RevlogError {
     fn corrupted() -> Self {
         RevlogError::Other(HgError::corrupted("corrupted revlog"))
     }
 }
 
 /// Read only implementation of revlog.
 pub struct Revlog {
     /// When index and data are not interleaved: bytes of the revlog index.
     /// When index and data are interleaved: bytes of the revlog index and
     /// data.
     index: Index,
     /// When index and data are not interleaved: bytes of the revlog data
     data_bytes: Option<Box<dyn Deref<Target = [u8]> + Send>>,
     /// When present on disk: the persistent nodemap for this revlog
     nodemap: Option<nodemap::NodeTree>,
 }
 
 impl Revlog {
     /// Open a revlog index file.
     ///
     /// It will also open the associated data file if index and data are not
     /// interleaved.
     #[timed]
     pub fn open(
         repo: &Repo,
         index_path: impl AsRef<Path>,
         data_path: Option<&Path>,
     ) -> Result<Self, HgError> {
         let index_path = index_path.as_ref();
         let index_mmap = repo.store_vfs().mmap_open(&index_path)?;
 
-        let version = get_version(&index_mmap);
+        let version = get_version(&index_mmap)?;
         if version != 1 {
             // A proper new version should have had a repo/store requirement.
             return Err(HgError::corrupted("corrupted revlog"));
         }
 
         let index = Index::new(Box::new(index_mmap))?;
 
         let default_data_path = index_path.with_extension("d");
 
         // type annotation required
         // won't recognize Mmap as Deref<Target = [u8]>
         let data_bytes: Option<Box<dyn Deref<Target = [u8]> + Send>> =
             if index.is_inline() {
                 None
             } else {
                 let data_path = data_path.unwrap_or(&default_data_path);
                 let data_mmap = repo.store_vfs().mmap_open(data_path)?;
                 Some(Box::new(data_mmap))
             };
 
         let nodemap = NodeMapDocket::read_from_file(repo, index_path)?.map(
             |(docket, data)| {
                 nodemap::NodeTree::load_bytes(
                     Box::new(data),
                     docket.data_length,
                 )
             },
         );
 
         Ok(Revlog {
             index,
             data_bytes,
             nodemap,
         })
     }
 
     /// Return number of entries of the `Revlog`.
     pub fn len(&self) -> usize {
         self.index.len()
     }
 
     /// Returns `true` if the `Revlog` has zero `entries`.
     pub fn is_empty(&self) -> bool {
         self.index.is_empty()
     }
 
     /// Returns the node ID for the given revision number, if it exists in this
     /// revlog
     pub fn node_from_rev(&self, rev: Revision) -> Option<&Node> {
         Some(self.index.get_entry(rev)?.hash())
     }
 
     /// Return the revision number for the given node ID, if it exists in this
     /// revlog
     #[timed]
     pub fn rev_from_node(
         &self,
         node: NodePrefix,
     ) -> Result<Revision, RevlogError> {
         if node.is_prefix_of(&NULL_NODE) {
             return Ok(NULL_REVISION);
         }
 
         if let Some(nodemap) = &self.nodemap {
             return nodemap
                 .find_bin(&self.index, node)?
                 .ok_or(RevlogError::InvalidRevision);
         }
 
         // Fallback to linear scan when a persistent nodemap is not present.
         // This happens when the persistent-nodemap experimental feature is not
         // enabled, or for small revlogs.
         //
         // TODO: consider building a non-persistent nodemap in memory to
         // optimize these cases.
         let mut found_by_prefix = None;
         for rev in (0..self.len() as Revision).rev() {
             let index_entry =
                 self.index.get_entry(rev).ok_or(HgError::corrupted(
                     "revlog references a revision not in the index",
                 ))?;
             if node == *index_entry.hash() {
                 return Ok(rev);
             }
             if node.is_prefix_of(index_entry.hash()) {
                 if found_by_prefix.is_some() {
                     return Err(RevlogError::AmbiguousPrefix);
                 }
                 found_by_prefix = Some(rev)
             }
         }
         found_by_prefix.ok_or(RevlogError::InvalidRevision)
     }
 
     /// Returns whether the given revision exists in this revlog.
     pub fn has_rev(&self, rev: Revision) -> bool {
         self.index.get_entry(rev).is_some()
     }
 
     /// Return the full data associated to a revision.
     ///
     /// All entries required to build the final data out of deltas will be
     /// retrieved as needed, and the deltas will be applied to the inital
     /// snapshot to rebuild the final data.
     #[timed]
     pub fn get_rev_data(&self, rev: Revision) -> Result<Vec<u8>, RevlogError> {
+        if rev == NULL_REVISION {
+            return Ok(vec![]);
+        };
         // Todo return -> Cow
         let mut entry = self.get_entry(rev)?;
         let mut delta_chain = vec![];
         while let Some(base_rev) = entry.base_rev {
             delta_chain.push(entry);
             entry = self
                 .get_entry(base_rev)
                 .map_err(|_| RevlogError::corrupted())?;
         }
 
         // TODO do not look twice in the index
         let index_entry = self
             .index
             .get_entry(rev)
             .ok_or(RevlogError::InvalidRevision)?;
 
         let data: Vec<u8> = if delta_chain.is_empty() {
             entry.data()?.into()
         } else {
             Revlog::build_data_from_deltas(entry, &delta_chain)?
         };
 
         if self.check_hash(
             index_entry.p1(),
             index_entry.p2(),
             index_entry.hash().as_bytes(),
             &data,
         ) {
             Ok(data)
         } else {
             Err(RevlogError::corrupted())
         }
     }
 
     /// Check the hash of some given data against the recorded hash.
     pub fn check_hash(
         &self,
         p1: Revision,
         p2: Revision,
         expected: &[u8],
         data: &[u8],
     ) -> bool {
         let e1 = self.index.get_entry(p1);
         let h1 = match e1 {
             Some(ref entry) => entry.hash(),
             None => &NULL_NODE,
         };
         let e2 = self.index.get_entry(p2);
         let h2 = match e2 {
             Some(ref entry) => entry.hash(),
             None => &NULL_NODE,
         };
 
         &hash(data, h1.as_bytes(), h2.as_bytes()) == expected
     }
 
     /// Build the full data of a revision out its snapshot
     /// and its deltas.
     #[timed]
     fn build_data_from_deltas(
         snapshot: RevlogEntry,
         deltas: &[RevlogEntry],
     ) -> Result<Vec<u8>, RevlogError> {
         let snapshot = snapshot.data()?;
         let deltas = deltas
             .iter()
             .rev()
             .map(RevlogEntry::data)
             .collect::<Result<Vec<Cow<'_, [u8]>>, RevlogError>>()?;
         let patches: Vec<_> =
             deltas.iter().map(|d| patch::PatchList::new(d)).collect();
         let patch = patch::fold_patch_lists(&patches);
         Ok(patch.apply(&snapshot))
     }
 
     /// Return the revlog data.
     fn data(&self) -> &[u8] {
         match self.data_bytes {
             Some(ref data_bytes) => &data_bytes,
             None => panic!(
                 "forgot to load the data or trying to access inline data"
             ),
         }
     }
 
     /// Get an entry of the revlog.
     fn get_entry(&self, rev: Revision) -> Result<RevlogEntry, RevlogError> {
         let index_entry = self
             .index
             .get_entry(rev)
             .ok_or(RevlogError::InvalidRevision)?;
         let start = index_entry.offset();
         let end = start + index_entry.compressed_len();
         let data = if self.index.is_inline() {
             self.index.data(start, end)
         } else {
             &self.data()[start..end]
         };
         let entry = RevlogEntry {
             rev,
             bytes: data,
             compressed_len: index_entry.compressed_len(),
             uncompressed_len: index_entry.uncompressed_len(),
             base_rev: if index_entry.base_revision() == rev {
                 None
             } else {
                 Some(index_entry.base_revision())
             },
         };
         Ok(entry)
     }
 }
 
 /// The revlog entry's bytes and the necessary informations to extract
 /// the entry's data.
 #[derive(Debug)]
 pub struct RevlogEntry<'a> {
     rev: Revision,
     bytes: &'a [u8],
     compressed_len: usize,
     uncompressed_len: usize,
     base_rev: Option<Revision>,
 }
 
 impl<'a> RevlogEntry<'a> {
     pub fn revision(&self) -> Revision {
         self.rev
     }
 
     /// Extract the data contained in the entry.
     pub fn data(&self) -> Result<Cow<'_, [u8]>, RevlogError> {
         if self.bytes.is_empty() {
             return Ok(Cow::Borrowed(&[]));
         }
         match self.bytes[0] {
             // Revision data is the entirety of the entry, including this
             // header.
             b'\0' => Ok(Cow::Borrowed(self.bytes)),
             // Raw revision data follows.
             b'u' => Ok(Cow::Borrowed(&self.bytes[1..])),
             // zlib (RFC 1950) data.
             b'x' => Ok(Cow::Owned(self.uncompressed_zlib_data()?)),
             // zstd data.
             b'\x28' => Ok(Cow::Owned(self.uncompressed_zstd_data()?)),
             // A proper new format should have had a repo/store requirement.
             _format_type => Err(RevlogError::corrupted()),
         }
     }
 
     fn uncompressed_zlib_data(&self) -> Result<Vec<u8>, RevlogError> {
         let mut decoder = ZlibDecoder::new(self.bytes);
         if self.is_delta() {
             let mut buf = Vec::with_capacity(self.compressed_len);
             decoder
                 .read_to_end(&mut buf)
                 .map_err(|_| RevlogError::corrupted())?;
             Ok(buf)
         } else {
             let mut buf = vec![0; self.uncompressed_len];
             decoder
                 .read_exact(&mut buf)
                 .map_err(|_| RevlogError::corrupted())?;
             Ok(buf)
         }
     }
 
     fn uncompressed_zstd_data(&self) -> Result<Vec<u8>, RevlogError> {
         if self.is_delta() {
             let mut buf = Vec::with_capacity(self.compressed_len);
             zstd::stream::copy_decode(self.bytes, &mut buf)
                 .map_err(|_| RevlogError::corrupted())?;
             Ok(buf)
         } else {
             let mut buf = vec![0; self.uncompressed_len];
             let len = zstd::block::decompress_to_buffer(self.bytes, &mut buf)
                 .map_err(|_| RevlogError::corrupted())?;
             if len != self.uncompressed_len {
                 Err(RevlogError::corrupted())
             } else {
                 Ok(buf)
             }
         }
     }
 
     /// Tell if the entry is a snapshot or a delta
     /// (influences on decompression).
     fn is_delta(&self) -> bool {
         self.base_rev.is_some()
     }
 }
 
 /// Format version of the revlog.
-pub fn get_version(index_bytes: &[u8]) -> u16 {
-    BigEndian::read_u16(&index_bytes[2..=3])
+pub fn get_version(index_bytes: &[u8]) -> Result<u16, HgError> {
+    if index_bytes.len() == 0 {
+        return Ok(1);
+    };
+    if index_bytes.len() < 4 {
+        return Err(HgError::corrupted(
+            "corrupted revlog: can't read the index format header",
+        ));
+    };
+    Ok(BigEndian::read_u16(&index_bytes[2..=3]))
 }
 
 /// Calculate the hash of a revision given its data and its parents.
 fn hash(
     data: &[u8],
     p1_hash: &[u8],
     p2_hash: &[u8],
 ) -> [u8; NODE_BYTES_LENGTH] {
     let mut hasher = Sha1::new();
     let (a, b) = (p1_hash, p2_hash);
     if a > b {
         hasher.update(b);
         hasher.update(a);
     } else {
         hasher.update(a);
         hasher.update(b);
     }
     hasher.update(data);
     *hasher.finalize().as_ref()
 }
 
 #[cfg(test)]
 mod tests {
     use super::*;
 
     use super::super::index::IndexEntryBuilder;
 
     #[test]
     fn version_test() {
         let bytes = IndexEntryBuilder::new()
             .is_first(true)
             .with_version(1)
             .build();
 
         assert_eq!(get_version(&bytes), 1)
     }
 }