[ngrams] ignore errors when parsing TSV files

Related to
#433

Textual formats often have either encoding or column errors.

This is a common issue, see e.g.
https://duckdb.org/docs/data/csv/reading_faulty_csv_files.html

We just ignore them.

gargantext.cabal

@@ -32,6 +32,7 @@ data-files:
     ekg-assets/bootstrap-1.4.0.min.css
     ekg-assets/chart_line_add.png
     ekg-assets/cross.png
+    test-data/ngrams/433-utf-encoding-issue.tsv
     test-data/ngrams/GarganText_DocsList-nodeId-177.json
     test-data/ngrams/GarganText_NgramsTerms-nodeId-177.json
     test-data/ngrams/GarganText_NgramsTerms-QuantumComputing.json

src/Gargantext/API/Ngrams/List.hs

@@ -171,27 +171,28 @@ ngramsListFromTSVData tsvData = case decodeTsv of
     binaryData = BSL.fromStrict $ P.encodeUtf8 tsvData
 
     decodeTsv :: Either Prelude.String (Vector NgramsTableMap)
-    decodeTsv = Tsv.decodeWithP tsvToNgramsTableMap
+    decodeTsv = Vec.catMaybes <$>
+                   Tsv.decodeWithP tsvToNgramsTableMap
                                    (Tsv.defaultDecodeOptions { Tsv.decDelimiter = fromIntegral (P.ord '\t') })
                                    Tsv.HasHeader
                                    binaryData
 
 -- | Converts a plain TSV 'Record' into an NgramsTableMap
-tsvToNgramsTableMap :: Tsv.Record -> Tsv.Parser NgramsTableMap
+tsvToNgramsTableMap :: Tsv.Record -> Tsv.Parser (Maybe NgramsTableMap)
 tsvToNgramsTableMap record = case Vec.toList record of
   (map P.decodeUtf8 -> [status, label, forms])
-    -> pure $ conv status label forms
-  _ -> Prelude.fail "tsvToNgramsTableMap failed"
+    -> pure $ Just $ conv status label forms
+  -- WARNING: This silently ignores errors (#433)
+  _ -> pure Nothing
 
   where
     conv :: Text -> Text -> Text -> NgramsTableMap
     conv status label forms = Map.singleton (NgramsTerm label)
         $ NgramsRepoElement { _nre_size = 1
-                                             , _nre_list = case status == "map" of
-                                                             True  -> MapTerm
-                                                             False -> case status == "main" of
-                                                                True  -> CandidateTerm
-                                                                False -> StopTerm
+                            , _nre_list = case status of
+                                            "map"  -> MapTerm
+                                            "main" -> CandidateTerm
+                                            _      -> StopTerm
                             , _nre_root = Nothing
                             , _nre_parent = Nothing
                             , _nre_children = MSet

test/Test/API/UpdateList.hs

@@ -209,6 +209,11 @@ tests = sequential $ aroundAll withTestDBAndPort $ beforeAllWith dbEnvSetup $ do
                                 , (NgramsTerm "brazorf", NgramsRepoElement 1 StopTerm Nothing Nothing (MSet mempty))
                               ])])
 
+      it "parses TSV with UTF-8 issues" $ \(SpecContext _testEnv _port _app _) -> do
+        simpleNgrams <- liftIO (TIO.readFile =<< getDataFileName "test-data/ngrams/433-utf-encoding-issue.tsv")
+        -- we don't care about the output, only that the file was parsed without errors (this file is garbage)
+        ngramsListFromTSVData simpleNgrams `shouldSatisfy` isRight
+
       it "allows uploading a CSV ngrams file" $ \(SpecContext testEnv port app _) -> do
         cId <- newCorpusForUser testEnv "alice"
         withApplication app $ do