add version check in job script (#206)

Author: rvhonorato

src/checksum.rs

@@ -1,4 +1,4 @@
-use anyhow::{Context, Result, bail};
+use anyhow::{bail, Context, Result};
 use md5::{Digest, Md5};
 use serde::{Deserialize, Serialize};
 use std::collections::HashMap;
@@ -175,10 +175,7 @@ pub fn validate_checksums(
 ///
 /// * `Result<()>` - Ok if the live version matches the stored one, Err otherwise
 pub fn validate_haddock3_version_live(checksum_file: &Path) -> Result<()> {
-    let stored_content =
-        fs::read_to_string(checksum_file).context("Failed to read checksum file")?;
-    let stored_data: ChecksumData =
-        serde_json::from_str(&stored_content).context("Failed to parse checksum file")?;
+    let stored_data = read_checksum_data(checksum_file)?;
 
     let live_version = crate::utils::get_haddock3_version()?;
 
@@ -193,6 +190,20 @@ pub fn validate_haddock3_version_live(checksum_file: &Path) -> Result<()> {
     Ok(())
 }
 
+fn read_checksum_data(checksum_file: &Path) -> Result<ChecksumData> {
+    let stored_content =
+        fs::read_to_string(checksum_file).context("Failed to read checksum file")?;
+    serde_json::from_str(&stored_content).context("Failed to parse checksum file")
+}
+
+/// Returns the haddock3 version recorded in `checksum.json` when the benchmark started.
+///
+/// Used to embed the expected version into a generated SLURM job script so it can be
+/// re-checked from inside the job itself, after `slurm_prologue` has run.
+pub fn expected_haddock3_version(checksum_file: &Path) -> Result<String> {
+    Ok(read_checksum_data(checksum_file)?.haddock3_version)
+}
+
 fn find_modified(
     current_checksums: HashMap<String, String>,
     stored_checksums: HashMap<String, String>,
@@ -265,7 +276,7 @@ mod tests {
     use super::*;
     use std::collections::HashMap;
     use std::io::Write;
-    use tempfile::{NamedTempFile, tempdir};
+    use tempfile::{tempdir, NamedTempFile};
 
     #[test]
     fn test_calculate_checksum() {
@@ -433,7 +444,7 @@ mod tests {
         let mut stored = HashMap::new();
         stored.insert("file1.txt".to_string(), "checksum1_changed".to_string()); // Changed
         stored.insert("file4.txt".to_string(), "checksum4".to_string()); // Removed
-        // file2.txt and file3.txt are new
+                                                                         // file2.txt and file3.txt are new
 
         // Find modified files
         let error_msg = find_modified(current, stored);
@@ -570,4 +581,19 @@ mod tests {
         let result = validate_haddock3_version_live(&checksum_file);
         assert!(result.is_err());
     }
+
+    #[test]
+    fn test_expected_haddock3_version() {
+        let temp_dir = tempdir().unwrap();
+        let checksum_file = temp_dir.path().join("checksum.json");
+
+        let data = ChecksumData {
+            files: HashMap::new(),
+            haddock3_version: "2026.3.0".to_string(),
+        };
+        fs::write(&checksum_file, serde_json::to_string_pretty(&data).unwrap()).unwrap();
+
+        let version = expected_haddock3_version(&checksum_file).unwrap();
+        assert_eq!(version, "2026.3.0");
+    }
 }

src/job.rs

@@ -405,22 +405,34 @@ impl Job {
         let mut file = fs::File::create(&job_script)?;
 
         let absolute_wd = canonicalize(&self.wd).unwrap_or_else(|_| self.wd.to_path_buf());
+        let haddock3_exe = find_haddock3_executable()?;
 
         // Write SLURM header
         let header = self.generate_slurm_header();
 
         let prologue = self.format_slurm_prologue();
 
+        // slurm_prologue can alter the environment (module load/conda activate/etc.)
+        // before haddock3 runs; re-check the version from inside the job itself so
+        // a swap isn't invisible to the login-node-only checks in checksum.rs.
+        let mut version_check = String::new();
+        if self.general.slurm_prologue.is_some() {
+            let checksum_file = self.general.work_dir.join("checksum.json");
+            let expected_version = crate::checksum::expected_haddock3_version(&checksum_file)?;
+            version_check = format_haddock3_version_check(&haddock3_exe, &expected_version);
+        }
+
         // Write job body
         let body = format!(
             "cd {}\n{} {}\n",
             absolute_wd.display(),
-            find_haddock3_executable()?,
+            haddock3_exe,
             WORKFLOW_FILENAME
         );
 
         file.write_all(header.as_bytes())?;
         file.write_all(prologue.as_bytes())?;
+        file.write_all(version_check.as_bytes())?;
         file.write_all(body.as_bytes())?;
 
         Ok(())
@@ -540,6 +552,21 @@ impl Job {
     }
 }
 
+/// Renders a shell snippet that re-checks `haddock3_exe --version` against `expected_version`,
+/// exiting the job script with an error on mismatch. Meant to run after `slurm_prologue`
+/// commands, in the same shell state, since those can alter the environment haddock3 sees.
+fn format_haddock3_version_check(haddock3_exe: &str, expected_version: &str) -> String {
+    format!(
+        r#"HADDOCK3_VERSION_OUTPUT="$({haddock3_exe} --version 2>&1)"
+ACTUAL_HADDOCK3_VERSION="${{HADDOCK3_VERSION_OUTPUT##* - }}"
+if [ "$ACTUAL_HADDOCK3_VERSION" != "{expected_version}" ]; then
+    echo "ERROR: haddock3 version mismatch after slurm_prologue: expected {expected_version}, found $ACTUAL_HADDOCK3_VERSION. Check slurm_prologue for environment changes (module load/conda activate/etc.)." >&2
+    exit 1
+fi
+"#
+    )
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -1073,6 +1100,15 @@ mod tests {
         assert_eq!(job.format_slurm_prologue(), "module load foo\n");
     }
 
+    #[test]
+    fn test_format_haddock3_version_check_contents() {
+        let check = format_haddock3_version_check("/opt/haddock3/bin/haddock3", "2026.3.0");
+        assert!(check.contains("/opt/haddock3/bin/haddock3 --version"));
+        assert!(check.contains("2026.3.0"));
+        assert!(check.contains("ACTUAL_HADDOCK3_VERSION"));
+        assert!(check.contains("exit 1"));
+    }
+
     fn make_job_with_modules(modules: IndexMap<String, Value>) -> Job {
         Job {
             name: "test".to_string(),