From 2c33a4f66a9b54a3f2ecbacf1356571f5099edbc Mon Sep 17 00:00:00 2001
From: Kriday Dave <technocratix902@gmail.com>
Date: Sun, 8 Nov 2026 14:51:10 +0530
Subject: [PATCH] fix: reject non-UTF-8 git paths to avoid snapshot collisions
 (SEC-10)

---
 src/adapter/git.rs        | 54 ++++++++++++++++++++++-----------------
 tests/git_adapter_test.rs | 25 ++++++++++++++++++
 2 files changed, 56 insertions(+), 23 deletions(-)

diff --git a/src/adapter/git.rs b/src/adapter/git.rs
index 3740886..dfece4e 100644
--- a/src/adapter/git.rs
+++ b/src/adapter/git.rs
@@ -158,30 +158,38 @@ impl GitAdapter {
             if entry.is_empty() {
                 continue;
             }
-            let entry_str = String::from_utf8_lossy(entry);
-            if let Some((meta, path)) = entry_str.split_once('\t') {
-                let parts: Vec<&str> = meta.split_whitespace().collect();
-                if parts.len() >= 3 && parts[1] == "blob" {
-                    // Skip symlinks (mode 120000)
-                    if parts[0] == "120000" {
-                        continue;
-                    }
-                    crate::store::validate_tree_path(path)?;
-                    let blob_sha = parts[2];
-                    let blob_output = Command::new("git")
-                        .args(["cat-file", "blob", blob_sha])
-                        .current_dir(&self.repo_root)
-                        .output()
-                        .with_context(|| format!("Failed to fetch blob {blob_sha} for {path}"))?;
-
-                    if !blob_output.status.success() {
-                        return Err(anyhow!(
-                            "Failed to read blob {blob_sha} for {path}: {}",
-                            String::from_utf8_lossy(&blob_output.stderr).trim()
-                        ));
-                    }
-                    files.insert(path.to_string(), blob_output.stdout);
+            let Some(tab) = entry.iter().position(|&b| b == b'\t') else {
+                continue;
+            };
+            let meta = String::from_utf8_lossy(&entry[..tab]);
+            let path_bytes = &entry[tab + 1..];
+            let path = std::str::from_utf8(path_bytes).map_err(|_| {
+                anyhow!(
+                    "git tree contains a non-UTF-8 path in '{rev}': {} (rejecting to avoid path collisions)",
+                    String::from_utf8_lossy(path_bytes)
+                )
+            })?;
+            let parts: Vec<&str> = meta.split_whitespace().collect();
+            if parts.len() >= 3 && parts[1] == "blob" {
+                // Skip symlinks (mode 120000)
+                if parts[0] == "120000" {
+                    continue;
+                }
+                crate::store::validate_tree_path(path)?;
+                let blob_sha = parts[2];
+                let blob_output = Command::new("git")
+                    .args(["cat-file", "blob", blob_sha])
+                    .current_dir(&self.repo_root)
+                    .output()
+                    .with_context(|| format!("Failed to fetch blob {blob_sha} for {path}"))?;
+
+                if !blob_output.status.success() {
+                    return Err(anyhow!(
+                        "Failed to read blob {blob_sha} for {path}: {}",
+                        String::from_utf8_lossy(&blob_output.stderr).trim()
+                    ));
                 }
+                files.insert(path.to_string(), blob_output.stdout);
             }
         }
 
diff --git a/tests/git_adapter_test.rs b/tests/git_adapter_test.rs
index ffe45ed..2949160 100644
--- a/tests/git_adapter_test.rs
+++ b/tests/git_adapter_test.rs
@@ -507,3 +507,28 @@ fn test_cli_git_adjudication_respects_authors_flag() {
         "VCS adjudication must honor --authors; stdout was: {stdout}"
     );
 }
+
+#[cfg(unix)]
+#[test]
+fn test_git_adapter_rejects_non_utf8_path() {
+    use std::ffi::OsString;
+    use std::os::unix::ffi::OsStringExt;
+
+    let repo = TempGitRepo::new("non_utf8_path");
+    repo.write_file("src/lib.rs", "pub fn ok() {}\n");
+    // Invalid UTF-8 in the filename: lossy decoding would collapse distinct
+    // byte paths to the same key (SEC-10), so extraction must reject it loudly.
+    let bad_name = OsString::from_vec(b"bad\xff.txt".to_vec());
+    fs::write(repo.path.join(&bad_name), b"secret").expect("failed to write non-utf8 file");
+    let rev = repo.commit("add non-utf8 path");
+
+    let adapter = GitAdapter::new(&repo.path).expect("valid git repo");
+    let err = adapter
+        .extract_snapshot(&rev)
+        .expect_err("non-UTF-8 paths must not be lossily decoded into snapshot keys");
+    let msg = err.to_string();
+    assert!(
+        msg.contains("non-UTF-8 path"),
+        "error should identify the non-UTF-8 path, got: {msg}"
+    );
+}