fix(builtins): preserve raw bytes from /dev/urandom through pipeline (#870)

Closes #811

Author: chaliy

crates/bashkit/src/builtins/cuttr.rs

@@ -319,10 +319,12 @@ impl Builtin for Tr {
 
         let mut set1 = expand_char_set(non_flag_args[0]);
         if complement {
-            // Complement: use all ASCII chars NOT in set1
+            // Complement: use all byte-range chars (0-255) NOT in set1.
+            // Covers full Latin-1 range so binary data from /dev/urandom
+            // (where each byte maps to one char) is handled correctly.
             let original = set1.clone();
-            set1 = (0u8..=127)
-                .map(|b| b as char)
+            set1 = (0u16..=255)
+                .map(|b| b as u8 as char)
                 .filter(|c| !original.contains(c))
                 .collect();
         }

crates/bashkit/src/builtins/headtail.rs

@@ -165,11 +165,12 @@ fn parse_head_args(args: &[String], default: usize) -> Result<(usize, bool, Vec<
     Ok((count, byte_mode, files))
 }
 
-/// Take the first N bytes from text
+/// Take the first N bytes from text.
+/// Uses char-level truncation so that Latin-1 encoded binary data
+/// (e.g. from /dev/urandom where each byte maps to one char) is
+/// counted correctly — each char represents one original byte.
 fn take_first_bytes(text: &str, n: usize) -> String {
-    let bytes = text.as_bytes();
-    let take = bytes.len().min(n);
-    String::from_utf8_lossy(&bytes[..take]).to_string()
+    text.chars().take(n).collect()
 }
 
 /// Parse arguments for tail command, including +N "from start" syntax.

crates/bashkit/src/builtins/mod.rs

@@ -220,6 +220,13 @@ pub(crate) async fn read_text_file(
         .await
         .map_err(|e| ExecResult::err(format!("{cmd_name}: {}: {e}\n", path.display()), 1))?;
 
+    // Binary device files (/dev/urandom, /dev/random): preserve raw bytes as
+    // Latin-1 (ISO 8859-1) so each byte 0x00-0xFF maps 1:1 to a char.
+    // This lets `tr -dc 'a-z0-9' < /dev/urandom | head -c N` work correctly.
+    if path == Path::new("/dev/urandom") || path == Path::new("/dev/random") {
+        return Ok(content.iter().map(|&b| b as char).collect());
+    }
+
     Ok(String::from_utf8_lossy(&content).into_owned())
 }
 

crates/bashkit/tests/urandom_tests.rs

@@ -19,3 +19,46 @@ async fn urandom_no_replacement_chars() {
         &hex[..hex.len().min(60)]
     );
 }
+
+/// Issue #811: head -c N /dev/urandom should return exactly N chars
+/// (each original byte maps to one char in the Latin-1 model)
+#[tokio::test]
+async fn urandom_head_char_count() {
+    let mut bash = Bash::new();
+    for n in [1, 4, 8, 16, 32] {
+        let result = bash
+            .exec(&format!("head -c {n} /dev/urandom | wc -m"))
+            .await
+            .unwrap();
+        let count: usize = result.stdout.trim().parse().unwrap_or(0);
+        assert_eq!(
+            count, n,
+            "head -c {n} /dev/urandom | wc -m should produce exactly {n} chars"
+        );
+    }
+}
+
+/// Issue #811: tr -dc 'a-z0-9' < /dev/urandom | head -c 8 should produce 8 alphanumeric chars
+#[tokio::test]
+async fn urandom_tr_filter_alphanumeric() {
+    let mut bash = Bash::new();
+    let result = bash
+        .exec("LC_ALL=C tr -dc 'a-z0-9' < /dev/urandom | head -c 8")
+        .await
+        .unwrap();
+    let output = result.stdout.trim();
+    assert_eq!(
+        output.len(),
+        8,
+        "Should produce exactly 8 chars, got {}: {:?}",
+        output.len(),
+        output
+    );
+    assert!(
+        output
+            .chars()
+            .all(|c| c.is_ascii_lowercase() || c.is_ascii_digit()),
+        "All chars should be a-z0-9, got: {:?}",
+        output
+    );
+}