Support VariableList longer than 2**31 on 32-bit architectures

* Improves handling large VariableList for 32-bit architectures
   - loudly crash vs. silently overflow (and produce wrong results)
* Adds feature to enable capping typenum to usize conversion to usize::MAX
* Tests + github actions

Author: color-typea

.github/workflows/test-suite.yml

@@ -31,6 +31,28 @@ jobs:
       run: rustup update stable
     - name: Run tests
       run: cargo test --release
+  cross-test-i686:
+    name: cross test i686-unknown-linux-gnu
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v3
+    - name: Install cross
+      run: cargo install cross --git https://github.com/cross-rs/cross
+    - name: Add i686-unknown-linux-gnu target
+      run: rustup target add i686-unknown-linux-gnu
+    - name: Run cross test for i686-unknown-linux-gnu
+      run: cross test --target i686-unknown-linux-gnu
+  cross-test-i686-overflow:
+    name: cross test i686-unknown-linux-gnu (typenum overflow feature)
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v3
+    - name: Install cross
+      run: cargo install cross --git https://github.com/cross-rs/cross
+    - name: Add i686-unknown-linux-gnu target
+      run: rustup target add i686-unknown-linux-gnu
+    - name: Run cross test for i686-unknown-linux-gnu with cap-typenum-to-usize-overflow
+      run: cross test --target i686-unknown-linux-gnu --features cap-typenum-to-usize-overflow
   coverage:
     name: cargo-tarpaulin
     runs-on: ubuntu-latest

Cargo.toml

@@ -20,7 +20,16 @@ typenum = "1.12.0"
 smallvec = "1.8.0"
 arbitrary = { version = "1.0", features = ["derive"], optional = true }
 itertools = "0.13.0"
+ethereum_hashing = {version = "0.7.0", optional = true}
 
 [dev-dependencies]
 serde_json = "1.0.0"
 tree_hash_derive = "0.10.0"
+ethereum_hashing = {version = "0.7.0"}
+
+[target.i686-unknown-linux-gnu]
+rustflags = ["-C", "target-feature=+sse2"]
+
+[features]
+# Very careful usage - see comment in the typenum_helpers
+cap-typenum-to-usize-overflow=["dep:ethereum_hashing"]

src/lib.rs

@@ -42,6 +42,7 @@ mod fixed_vector;
 pub mod serde_utils;
 mod tree_hash;
 mod variable_list;
+mod typenum_helpers;
 
 pub use fixed_vector::FixedVector;
 pub use ssz::{BitList, BitVector, Bitfield};

src/tree_hash.rs

@@ -1,41 +1,96 @@
 use tree_hash::{Hash256, MerkleHasher, TreeHash, TreeHashType};
 use typenum::Unsigned;
+use crate::typenum_helpers::to_usize;
 
-/// A helper function providing common functionality between the `TreeHash` implementations for
-/// `FixedVector` and `VariableList`.
-pub fn vec_tree_hash_root<T, N>(vec: &[T]) -> Hash256
-where
-    T: TreeHash,
-    N: Unsigned,
-{
+pub fn packing_factor<T: TreeHash>() -> usize {
     match T::tree_hash_type() {
-        TreeHashType::Basic => {
-            let mut hasher = MerkleHasher::with_leaves(
-                (N::to_usize() + T::tree_hash_packing_factor() - 1) / T::tree_hash_packing_factor(),
-            );
+        TreeHashType::Basic => T::tree_hash_packing_factor(),
+        TreeHashType::Container | TreeHashType::List | TreeHashType::Vector => 1,
+    }
+}
+
+mod default_impl {
+    use super::*;
+    /// A helper function providing common functionality between the `TreeHash` implementations for
+    /// `FixedVector` and `VariableList`.
+    pub fn vec_tree_hash_root<T, N>(vec: &[T]) -> Hash256
+    where
+        T: TreeHash,
+        N: Unsigned,
+    {
+        match T::tree_hash_type() {
+            TreeHashType::Basic => {
+                let mut hasher = MerkleHasher::with_leaves(
+                    (to_usize::<N>() + T::tree_hash_packing_factor() - 1) / T::tree_hash_packing_factor(),
+                );
+
+                for item in vec {
+                    hasher
+                        .write(&item.tree_hash_packed_encoding())
+                        .expect("ssz_types variable vec should not contain more elements than max");
+                }
 
-            for item in vec {
                 hasher
-                    .write(&item.tree_hash_packed_encoding())
-                    .expect("ssz_types variable vec should not contain more elements than max");
+                    .finish()
+                    .expect("ssz_types variable vec should not have a remaining buffer")
             }
+            TreeHashType::Container | TreeHashType::List | TreeHashType::Vector => {
+                let mut hasher = MerkleHasher::with_leaves(N::to_usize());
 
-            hasher
-                .finish()
-                .expect("ssz_types variable vec should not have a remaining buffer")
-        }
-        TreeHashType::Container | TreeHashType::List | TreeHashType::Vector => {
-            let mut hasher = MerkleHasher::with_leaves(N::to_usize());
+                for item in vec {
+                    hasher
+                        .write(item.tree_hash_root().as_slice())
+                        .expect("ssz_types vec should not contain more elements than max");
+                }
 
-            for item in vec {
                 hasher
-                    .write(item.tree_hash_root().as_slice())
-                    .expect("ssz_types vec should not contain more elements than max");
+                    .finish()
+                    .expect("ssz_types vec should not have a remaining buffer")
             }
+        }
+    }
+}
+
+#[cfg(feature="cap-typenum-to-usize-overflow")]
+mod arch_32x_workaround {
+    use super::*;
+    use tree_hash::{Hash256, TreeHash};
+    use typenum::Unsigned;
+    use ethereum_hashing::{hash32_concat, ZERO_HASHES};
+
+    type MaxDepth = typenum::U536870912;
 
-            hasher
-                .finish()
-                .expect("ssz_types vec should not have a remaining buffer")
+    fn pad_to_depth<Current: Unsigned, Target: Unsigned>(hash: Hash256, target_depth: usize, current_depth: usize) -> Hash256 {
+        let mut curhash: [u8; 32] = hash.0;
+        for depth in current_depth..target_depth {
+            curhash = hash32_concat(&curhash, ZERO_HASHES[depth].as_slice());
+        }
+        curhash.into()
+    }
+
+    fn target_tree_depth<T: TreeHash, N: Unsigned>() -> usize {
+        let packing_factor = packing_factor::<T>();
+        let packing_factor_log2 = packing_factor.next_power_of_two().ilog2() as usize;
+        let tree_depth = N::to_u64().next_power_of_two().ilog2() as usize;
+        tree_depth - packing_factor_log2
+    }
+
+    pub fn vec_tree_hash_root<T: TreeHash, N: Unsigned>(vec: &[T]) -> Hash256 {
+        if N::to_u64() <= MaxDepth::to_u64() {
+            default_impl::vec_tree_hash_root::<T, N>(vec)
+        } else {
+            let main_tree_hash = default_impl::vec_tree_hash_root::<T, MaxDepth>(vec);
+
+            let target_depth = target_tree_depth::<T, N>();
+            let current_depth = target_tree_depth::<T, MaxDepth>();
+
+            pad_to_depth::<MaxDepth, N>(main_tree_hash, target_depth, current_depth)
         }
     }
 }
+
+#[cfg(any(target_pointer_width = "64", not(feature="cap-typenum-to-usize-overflow")))]
+pub use default_impl::vec_tree_hash_root;
+
+#[cfg(all(not(target_pointer_width = "64"), feature="cap-typenum-to-usize-overflow"))]
+pub use arch_32x_workaround::vec_tree_hash_root;

src/typenum_helpers.rs

@@ -0,0 +1,39 @@
+use typenum::Unsigned;
+
+// On x64, all typenums always fit usize
+#[cfg(target_pointer_width = "64")]
+pub fn to_usize<N:Unsigned>() -> usize {
+    N::to_usize()
+}
+
+
+// On x32, typenums larger starting from 2**32 do not fit usize,
+#[cfg(not(target_pointer_width = "64"))]
+pub fn to_usize<N:Unsigned>() -> usize {
+    let as_usize = N::to_usize();
+    let as_u64 = N::to_u64();
+    // If usize == u64 representation - N still fit usize, so 
+    // no overflow happened
+    if as_usize as u64 == as_u64 {
+        return as_usize;
+    }
+    // else we have a choice:
+    // Option 1. Loudly panic with as informative message as possible
+    #[cfg(not(feature="cap-typenum-to-usize-overflow"))]
+    panic!("Overflow converting typenum U{} to usize (usize::MAX={})", as_u64, usize::MAX);
+    // Option 2. Use usize::MAX - this allows working with VariableLists "virtually larger" than the 
+    // usize, provided the actual number of elements do not exceed usize.
+    //
+    // One example is Ethereum BeaconChain.validators field that is a VariableList<..., 2**40>, 
+    // but actual number of validators is far less than 2**32.
+    //
+    // This option still seems sound, since if the number of elements
+    // actually surpass usize::MAX, the machine running this will OOM/segfault/otherwise violently
+    // crash the program running this, which is nearly equivalent to panic.
+    //
+    // Still, the is a double-edged sword, only apply if you can guarantee that none of the
+    // VariableList used in your program will have more than usize::MAX elements on the
+    // architecture with the smallest usize it will be even run.
+    #[cfg(feature="cap-typenum-to-usize-overflow")]
+    usize::MAX
+}