Added Bloom filter verification

Commit-graph written with bloom filter helps accelerate many git
operations which require directory tree traversal to detect changed
paths.
However when a commit-graph chain were written, some parts of the chain
may not contains the bloom filter data for a set of included commits.
To overcome this, validate whether the Bloom filter was added
into the chain and potentially rebuild the entire chain if needed.

Bloom filter are optional chunks in commit-graph files: BIDX for Bloom
filter index and BDAT for Bloom filter data. [1]

To check for commit-graph existence in each graph file:
- Check whether the graph file contained optional chunks
- Among the optional chunks, Check for both BIDX and BDAT existence

Unless every graph file in the commit-graph chain contains Bloom filter,
rebuild the entire chain.

[1]: https://github.com/git/git/blob/ae46588be0cd730430dded4491246dfb4eac5557/Documentation/technical/commit-graph-format.txt#L99

Author: Son Luong Ngoc

git-care.sh

@@ -51,6 +51,62 @@ prefetch_loop() {
   done;
 }
 
+# verify_bloom_filter verify that bloom filter was used
+# in a split commit-graph chain. If not, remove the chain.
+verify_bloom_filter() {
+  for f in ${PROJECT_DIR}/.git/objects/info/commit-graphs/*.graph; do
+    chunk_count=$(od -j6 -N1 -An -i $f | tr -d '[:space:]')
+
+    # A commit graph always has 3 default chunks
+    # If there were less than 3 chunks, the file is corrupted
+    if [[ ${chunk_count} -lt 3 ]]; then
+      echo "Corrupted commit-graph in $f"
+      echo 'Rebuilding commit-graph chain'
+      rm -f ${PROJECT_DIR}/.git/objects/info/commit-graphs/commit-graph-chain;
+    fi
+
+    # If there were only 3 chunks, the Bloom filter chunks are missing
+    if [[ ${chunk_count} == 3 ]]; then
+      echo "Missing Bloom filter in $f"
+      echo 'Rebuilding commit-graph chain with bloom filter (slow)'
+      rm -f ${PROJECT_DIR}/.git/objects/info/commit-graphs/commit-graph-chain;
+    fi
+
+    has_bloom_index=0
+    has_bloom_data=0
+
+    # Find Bloom filter chunks among non-default chunks
+    # Always skip the first 3 default chunks
+    max_offset=$((chunk_count - 1))
+    for i in $(seq 3 ${max_offset}); do
+      # 8 bytes for commit-graph header
+      # 12 bytes for each chunk skipped
+      bytes_offset=$((8 + i * 12))
+
+      # skipped $bytes_offset bytes,
+      # read first 4 bytes of next chunk to get 4 characters chunk id
+      chunk_id=$(od -j${bytes_offset} -N4 -An -c $f | tr -d '[:space:]')
+      case $chunk_id in
+        'BIDX')
+          has_bloom_index=1
+          ;;
+        'BDAT')
+          has_bloom_data=1
+          ;;
+        *)
+          ;;
+      esac
+    done
+
+    # If either the Bloom filter chunks were missing
+    if [[ $((has_bloom_index + has_bloom_data)) != 2 ]]; then
+      echo "Missing Bloom filter in $f"
+      echo 'Rebuilding commit-graph chain with bloom filter (slow)'
+      rm -f ${PROJECT_DIR}/.git/objects/info/commit-graphs/commit-graph-chain;
+    fi
+  done
+}
+
 # commit_graph refreshes the commit-graph in a non-disruptive manner
 # thus speed up git operations over the commit history i.e. git-log
 commit_graph() {
@@ -62,6 +118,12 @@ commit_graph() {
     rm -f ${PROJECT_DIR}/.git/objects/info/commit-graph
   fi
 
+  # If a commit-graph chain already exists,
+  # check if every chain were built with Bloom filter.
+  if [ -f ${PROJECT_DIR}/.git/objects/info/commit-graphs/commit-graph-chain ]; then
+    verify_bloom_filter
+  fi
+
   git commit-graph write --reachable --split --changed-paths --size-multiple=4 --no-progress;
 
   if git commit-graph verify --shallow --no-progress; then