fix(interpreter): reset last_exit_code in VFS subprocess isolation (#850)

Subprocess isolation in execute_script_content did not reset
last_exit_code or nounset_error, causing VFS script subprocesses
to inherit stale parent state. In real bash, a subprocess starts
with $?=0. This caused set -euo pipefail scripts to see a non-zero
$? from the parent shell, leading to false errexit failures.

Closes #842

Author: chaliy

crates/bashkit/src/interpreter/mod.rs

@@ -4013,14 +4013,18 @@ impl Interpreter {
         let saved_aliases = self.aliases.clone();
         let saved_coproc = self.coproc_buffers.clone();
 
-        // Child only sees exported variables (env), not all shell variables
+        // Child only sees exported variables (env), not all shell variables.
+        // Reset last_exit_code so $? starts at 0 (matches real bash subprocess).
+        // Clear nounset_error to prevent parent expansion errors from leaking.
         self.variables = self.env.clone();
         self.arrays.clear();
         self.assoc_arrays.clear();
         self.functions.clear();
         self.traps.clear();
         self.aliases.clear();
         self.coproc_buffers.clear();
+        self.last_exit_code = 0;
+        self.nounset_error = None;
 
         // Push call frame: $0 = script name, $1..N = args
         self.call_stack = vec![CallFrame {

crates/bashkit/tests/script_execution_tests.rs

@@ -409,3 +409,92 @@ async fn declare_f_no_args_lists_all_functions() {
     assert!(result.stdout.contains("foo"));
     assert!(result.stdout.contains("bar"));
 }
+
+/// Issue #842: $? should be 0 at start of VFS script subprocess, even when
+/// parent had non-zero exit code. Subprocess isolation must reset last_exit_code
+/// to match real bash subprocess behavior.
+#[tokio::test]
+async fn exec_vfs_script_initial_exit_code_is_zero() {
+    let mut bash = Bash::new();
+    let fs = bash.fs();
+
+    fs.write_file(Path::new("/check_exit.sh"), b"#!/bin/bash\necho $?\n")
+        .await
+        .unwrap();
+    fs.chmod(Path::new("/check_exit.sh"), 0o755).await.unwrap();
+
+    // Parent has non-zero exit code, then run VFS script
+    let result = bash.exec("false; /check_exit.sh").await.unwrap();
+    assert_eq!(
+        result.stdout.trim(),
+        "0",
+        "VFS script subprocess should start with $?=0 (like real bash); got stdout={:?}",
+        result.stdout
+    );
+}
+
+/// Issue #842: `set -euo pipefail` in VFS script should not trigger false failure
+/// even after a prior non-zero exit code in the parent shell.
+#[tokio::test]
+async fn exec_vfs_script_set_e_after_prior_failure() {
+    let mut bash = Bash::new();
+    let fs = bash.fs();
+
+    fs.write_file(
+        Path::new("/test.sh"),
+        br#"#!/bin/bash
+set -euo pipefail
+MY_VAR="hello"
+echo "${MY_VAR}"
+"#,
+    )
+    .await
+    .unwrap();
+    fs.chmod(Path::new("/test.sh"), 0o755).await.unwrap();
+
+    let result = bash.exec("false; /test.sh").await.unwrap();
+    assert_eq!(
+        result.exit_code, 0,
+        "VFS script with set -e should succeed after prior failure; stdout={:?}, stderr={:?}",
+        result.stdout, result.stderr
+    );
+    assert_eq!(result.stdout.trim(), "hello");
+}
+
+/// Issue #842: Nested VFS scripts with set -e and command substitution.
+#[tokio::test]
+async fn exec_vfs_script_set_e_nested_scripts() {
+    let mut bash = Bash::new();
+    let fs = bash.fs();
+
+    fs.write_file(
+        Path::new("/inner.sh"),
+        br#"#!/bin/bash
+set -euo pipefail
+echo "inner_output"
+"#,
+    )
+    .await
+    .unwrap();
+    fs.chmod(Path::new("/inner.sh"), 0o755).await.unwrap();
+
+    fs.write_file(
+        Path::new("/outer.sh"),
+        br#"#!/bin/bash
+set -euo pipefail
+RESULT="$(/inner.sh)"
+echo "${RESULT}"
+"#,
+    )
+    .await
+    .unwrap();
+    fs.chmod(Path::new("/outer.sh"), 0o755).await.unwrap();
+
+    let result = bash.exec("/outer.sh").await.unwrap();
+    assert_eq!(
+        result.exit_code, 0,
+        "Nested VFS scripts with set -e should work; stdout={:?}, stderr={:?}",
+        result.stdout, result.stderr
+    );
+    assert_eq!(result.stdout.trim(), "inner_output");
+}

specs/006-threat-model.md

@@ -1207,6 +1207,7 @@ This section maps former vulnerability IDs to the new threat ID scheme and track
 | TM-ISO-021 | EXIT trap leaks across `exec()` calls | Cross-invocation interference — trap from exec N fires in exec N+1 | Reset traps in `reset_for_execution()` |
 | TM-ISO-022 | `$?` leaks across `exec()` calls | State pollution — exit code from previous exec visible to next exec | Reset `last_exit_code` in `reset_for_execution()` |
 | TM-ISO-023 | `set -e` leaks across `exec()` calls | Unexpected abort — shell options from previous exec affect next exec | Reset shell options in `reset_for_execution()` |
+| TM-ISO-024 | `$?` leaks into VFS subprocess | Parent `last_exit_code` visible inside VFS script subprocess, causing false `set -e` failures | Reset `last_exit_code = 0` and `nounset_error = None` in `execute_script_content` subprocess isolation |
 | TM-INT-007 | `/dev/urandom` empty with `head -c` | Weak randomness — `head -c 16 /dev/urandom` returns empty string | Fix virtual device pipe handling in head builtin |
 | TM-DOS-044 | Nested `$()` stack overflow (regression) | Process crash (SIGABRT) at depth ~50 despite #492 fix | Interpreter execution path may need separate depth tracking from lexer fix |
 

supply-chain/config.toml

@@ -207,7 +207,7 @@ version = "1.1.0"
 criteria = "safe-to-deploy"
 
 [[exemptions.cmake]]
-version = "0.1.57"
+version = "0.1.58"
 criteria = "safe-to-deploy"
 
 [[exemptions.cobs]]