From 483b9860b0d396a7058827dcfc78d414104e023a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Arma=C3=ABl=20Gu=C3=A9neau?= <armael.gueneau@ens-lyon.org>
Date: Wed, 18 Dec 2024 08:51:23 +0100
Subject: [PATCH] handle classification conflicts between concurrent users

---
 README.md   |  2 --
 src/main.rs | 23 ++++++++++++++++-------
 2 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/README.md b/README.md
index b047c38..01f1fd0 100644
--- a/README.md
+++ b/README.md
@@ -21,8 +21,6 @@
 - periodically refresh the database of users from forgejo, and merge them with
   the local db, handling updates in users data (triggering re-classification if
   needed)
-- properly handle concurrent clients of the webapp (and handle classification
-  conflicts)
 - add backend to store data on garage instead of local files
 - replate the `api_token` file with a better mechanism: oauth maybe?
 - improve error handling 
diff --git a/src/main.rs b/src/main.rs
index 69c05c3..7a6ffed 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -385,22 +385,31 @@ async fn load_db() -> anyhow::Result<(Db, Classifier)> {
 fn set_spam(db: &mut Db, classifier: &mut Classifier, ids: &[(UserId, bool)]) {
     eprintln!("updating classifier");
 
-    for (id, is_spam) in ids {
-        let tokens = db.tokens.get(id).unwrap();
+    for (user_id, is_spam) in ids {
+        // Train classifier with tokens from the user
+        let tokens = db.tokens.get(user_id).unwrap();
         if *is_spam {
             classifier.train_spam(tokens);
         } else {
             classifier.train_ham(tokens);
         }
+
+        match db.is_spam.get(user_id) {
+            Some(b) if b != is_spam => {
+                // classification conflict between concurrent queries.
+                // In this case we play it safe and erase the classification for this user;
+                // it will need to be manually classified again.
+                eprintln!("Classification conflict (uid %d), forget current user classification");
+                db.is_spam.remove(user_id);
+            },
+            _ => {
+                db.is_spam.insert(*user_id, *is_spam);
+            }
+        }
     }
 
     eprintln!("recomputing user scores");
     db.recompute_scores(&classifier);
-
-    eprintln!("updating db with new classification");
-    for (user_id, is_spam) in ids {
-        db.is_spam.insert(*user_id, *is_spam);
-    }
 }
 
 lazy_static! {