fix: track version in fingerprint dep-info files

Encodes the version information into Cargo's fingerprint dep-info files,
so that when the format encoding changes in the future,
Cargo understands a dep-info file was outdated and doesn't bother parsing it.

Since there was no version info encoded in the old format (call it v0),
to be compatible with older cargoes,
this PR works around it with a horrible hack.
It is explained in the doc comment of `EncodedDepInfo`.

Author: weihanglo

crates/cargo-test-support/src/lib.rs

@@ -1597,6 +1597,12 @@ pub fn assert_deps(project: &Project, fingerprint: &str, test_cb: impl Fn(&Path,
     assert!(files.next().is_none(), "expected only 1 dep-info file");
     let dep_info = fs::read(&info_path).unwrap();
     let dep_info = &mut &dep_info[..];
+
+    // Consume the magic marker and version. Here they don't really matter.
+    read_usize(dep_info);
+    read_u8(dep_info);
+    read_u8(dep_info);
+
     let deps = (0..read_usize(dep_info))
         .map(|_| {
             let ty = read_u8(dep_info);

src/cargo/core/compiler/fingerprint/dep_info.rs

@@ -22,6 +22,9 @@ use cargo_util::Sha256;
 use crate::CargoResult;
 use crate::CARGO_ENV;
 
+/// The current format version of [`EncodedDepInfo`].
+const CURRENT_ENCODED_DEP_INFO_VERSION: u8 = 1;
+
 /// The representation of the `.d` dep-info file generated by rustc
 #[derive(Default)]
 pub struct RustcDepInfo {
@@ -61,20 +64,35 @@ pub enum DepInfoPathType {
 /// Currently the format looks like:
 ///
 /// ```text
-/// +------------+------------+---------------+---------------+
-/// | # of files | file paths | # of env vars | env var pairs |
-/// +------------+------------+---------------+---------------+
+/// +--------+---------+------------+------------+---------------+---------------+
+/// | marker | version | # of files | file paths | # of env vars | env var pairs |
+/// +--------+---------+------------+------------+---------------+---------------+
 /// ```
 ///
 /// Each field represents
 ///
+/// * _Marker_ --- A magic marker to ensure older Cargoes that only recognize
+///   format v0 (prior to checksum support in [`f4ca7390`]) not go further
+///   parsing newer formats. Since [`EncodedDepInfo`] is just an optimization,
+///   to avoid adding any complexity, Cargo recognizes only one version of
+///   [`CURRENT_ENCODED_DEP_INFO_VERSION`].
+///   The current layout looks like
+///   ```text
+///   +----------------------------+
+///   | [0x01 0x00 0x00 0x00 0xff] |
+///   +----------------------------+
+///   ```
+///   These bytes will be interpreted as "one file tracked and invalid
+///   [`DepInfoPathType`] variant with 255" by older Cargoes, so they will just
+///   stop parsing. This could prevent some bad parsing in rust-lang/cargo#14712.
+/// * _Version_ --- The current format version.
 /// * _Number of files/envs_ --- A `u32` representing the number of things.
 /// * _File paths_ --- Zero or more paths of files the dep-info file depends on.
 ///   Each path is encoded as the following:
 ///
 ///   ```text
 ///   +-----------+-------------+------------+---------------+-----------+-------+
-///   | Path type | len of path | path bytes | cksum exists? | file size | cksum |
+///   | path type | len of path | path bytes | cksum exists? | file size | cksum |
 ///   +-----------+-------------+------------+---------------+-----------+-------+
 ///   ```
 /// * _Env var pairs_ --- Zero or more env vars the dep-info file depends on.
@@ -84,6 +102,8 @@ pub enum DepInfoPathType {
 ///   | len of key | key bytes | value exists? | len of value | value bytes |
 ///   +------------+-----------+---------------+--------------+-------------+
 ///   ```
+///
+/// [`f4ca7390`]: https://github.com/rust-lang/cargo/commit/f4ca739073185ea5e1148ff100bb4a06d3bf721d
 #[derive(Default, Debug, PartialEq, Eq)]
 pub struct EncodedDepInfo {
     pub files: Vec<(DepInfoPathType, PathBuf, Option<(u64, String)>)>,
@@ -93,6 +113,12 @@ pub struct EncodedDepInfo {
 impl EncodedDepInfo {
     pub fn parse(mut bytes: &[u8]) -> Option<EncodedDepInfo> {
         let bytes = &mut bytes;
+        read_magic_marker(bytes)?;
+        let version = read_u8(bytes)?;
+        if version != CURRENT_ENCODED_DEP_INFO_VERSION {
+            return None;
+        }
+
         let nfiles = read_usize(bytes)?;
         let mut files = Vec::with_capacity(nfiles);
         for _ in 0..nfiles {
@@ -129,6 +155,18 @@ impl EncodedDepInfo {
         }
         return Some(EncodedDepInfo { files, env });
 
+        /// See [`EncodedDepInfo`] for why a magic marker exists.
+        fn read_magic_marker(bytes: &mut &[u8]) -> Option<()> {
+            let _size = read_usize(bytes)?;
+            let path_type = read_u8(bytes)?;
+            if path_type != u8::MAX {
+                // Old depinfo. Give up parsing it.
+                return None;
+            } else {
+                Some(())
+            }
+        }
+
         fn read_usize(bytes: &mut &[u8]) -> Option<usize> {
             let ret = bytes.get(..4)?;
             *bytes = &bytes[4..];
@@ -162,6 +200,10 @@ impl EncodedDepInfo {
     pub fn serialize(&self) -> CargoResult<Vec<u8>> {
         let mut ret = Vec::new();
         let dst = &mut ret;
+
+        write_magic_marker(dst);
+        dst.push(CURRENT_ENCODED_DEP_INFO_VERSION);
+
         write_usize(dst, self.files.len());
         for (ty, file, checksum_info) in self.files.iter() {
             match ty {
@@ -189,6 +231,14 @@ impl EncodedDepInfo {
         }
         return Ok(ret);
 
+        /// See [`EncodedDepInfo`] for why a magic marker exists.
+        ///
+        /// There is an assumption that there is always at least a file.
+        fn write_magic_marker(dst: &mut Vec<u8>) {
+            write_usize(dst, 1);
+            dst.push(u8::MAX);
+        }
+
         fn write_bytes(dst: &mut Vec<u8>, val: impl AsRef<[u8]>) {
             let val = val.as_ref();
             write_usize(dst, val.len());
@@ -673,7 +723,7 @@ mod encoded_dep_info {
         #[rustfmt::skip]
         let data = [
             0x01, 0x00, 0x00, 0x00, 0xff,       // magic marker
-            0x01,                               // version
+            CURRENT_ENCODED_DEP_INFO_VERSION,   // version
             0x01, 0x00, 0x00, 0x00,             // # of files
             0x00,                               // path type
             0x04, 0x00, 0x00, 0x00,             // len of path
@@ -682,7 +732,13 @@ mod encoded_dep_info {
             0x00, 0x00, 0x00, 0x00,             // # of env vars
         ];
         // The current cargo doesn't recognize the magic marker.
-        assert!(EncodedDepInfo::parse(&data).is_none());
+        assert_eq!(
+            EncodedDepInfo::parse(&data).unwrap(),
+            EncodedDepInfo {
+                files: vec![(DepInfoPathType::PackageRootRelative, "rust".into(), None)],
+                env: Vec::new(),
+            }
+        );
     }
 
     #[test]