From 0b0702de396bc1de5a4af80169fa8c45c102d8e1 Mon Sep 17 00:00:00 2001
From: oabrivard <olivier@abrivard.fr>
Date: Mon, 23 Mar 2026 11:54:32 +0100
Subject: [PATCH] fix: strip null bytes from LLM output before saving to
 PostgreSQL JSONB

LLM output occasionally contains \u0000 null bytes (e.g., "annonc\u0000...")
which PostgreSQL rejects in JSONB columns. Added sanitize_json_null_bytes()
that recursively strips null bytes from all string values before DB insert.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 backend/src/services/synthesis.rs | 46 +++++++++++++++++++++++++++++++
 1 file changed, 46 insertions(+)

diff --git a/backend/src/services/synthesis.rs b/backend/src/services/synthesis.rs
index c38b41d..d95b95a 100644
--- a/backend/src/services/synthesis.rs
+++ b/backend/src/services/synthesis.rs
@@ -358,6 +358,10 @@ async fn run_generation_inner(
         AppError::Internal(anyhow::anyhow!("Failed to serialize sections: {}", e))
     })?;
 
+    // Strip \u0000 null bytes — LLM output occasionally contains them and
+    // PostgreSQL rejects them in JSONB columns.
+    let sections_json = sanitize_json_null_bytes(sections_json);
+
     let synthesis =
         db::syntheses::create(&state.pool, user_id, &week, &sections_json).await?;
 
@@ -368,6 +372,25 @@ async fn run_generation_inner(
 // Helper Functions
 // ───────────────────────────────────────────────────────────────────
 
+/// Recursively strip `\u0000` null bytes from JSON values.
+///
+/// PostgreSQL rejects null bytes in JSONB text. LLM output occasionally
+/// contains them (e.g., `Meta AI a annonc\u0000...`).
+fn sanitize_json_null_bytes(value: serde_json::Value) -> serde_json::Value {
+    match value {
+        serde_json::Value::String(s) => serde_json::Value::String(s.replace('\0', "")),
+        serde_json::Value::Array(arr) => {
+            serde_json::Value::Array(arr.into_iter().map(sanitize_json_null_bytes).collect())
+        }
+        serde_json::Value::Object(map) => serde_json::Value::Object(
+            map.into_iter()
+                .map(|(k, v)| (k, sanitize_json_null_bytes(v)))
+                .collect(),
+        ),
+        other => other,
+    }
+}
+
 /// Emit a progress event via the watch channel.
 fn emit_progress(tx: &watch::Sender<ProgressEvent>, step: &str, message: &str, percent: u8) {
     tx.send(ProgressEvent::Progress {
@@ -1265,4 +1288,27 @@ mod tests {
         let result = filter_homepage_urls(parsed);
         assert_eq!(result[0].1.len(), 1);
     }
+
+    #[test]
+    fn sanitize_null_bytes_in_json_strings() {
+        let json = serde_json::json!({
+            "title": "Hello\u{0000}World",
+            "items": [{"summary": "Text\u{0000}with\u{0000}nulls"}]
+        });
+        let sanitized = sanitize_json_null_bytes(json);
+        assert_eq!(sanitized["title"], "HelloWorld");
+        assert_eq!(sanitized["items"][0]["summary"], "Textwithnulls");
+    }
+
+    #[test]
+    fn sanitize_preserves_clean_json() {
+        let json = serde_json::json!({
+            "title": "Clean text",
+            "count": 42,
+            "active": true,
+            "items": [{"url": "https://example.com"}]
+        });
+        let sanitized = sanitize_json_null_bytes(json.clone());
+        assert_eq!(sanitized, json);
+    }
 }