Fixes the download and usage of the downloaded data

docker/docker-entrypoint.sh

@@ -53,6 +53,21 @@ map_folders() {
 	export CONSUME_DIR="${PAPERLESS_CONSUMPTION_DIR:-/usr/src/paperless/consume}"
 }
 
+nltk_data () {
+	# Store the NLTK data outside the Docker container
+	local nltk_data_dir="${DATA_DIR}/nltk"
+
+	# Download or update the snowball stemmer data
+	python3 -W ignore::RuntimeWarning -m nltk.downloader -d "${nltk_data_dir}" snowball_data
+
+	# Download or update the stopwords corpus
+	python3 -W ignore::RuntimeWarning -m nltk.downloader -d "${nltk_data_dir}" stopwords
+
+	# Download or update the punkt tokenizer data
+	python3 -W ignore::RuntimeWarning -m nltk.downloader -d "${nltk_data_dir}" punkt
+
+}
+
 initialize() {
 
 	# Setup environment from secrets before anything else
@@ -93,6 +108,8 @@ initialize() {
 	echo "Creating directory ${tmp_dir}"
 	mkdir -p "${tmp_dir}"
 
+	nltk_data
+
 	set +e
 	echo "Adjusting permissions of paperless files. This may take a while."
 	chown -R paperless:paperless ${tmp_dir}

docker/docker-prepare.sh

@@ -89,24 +89,6 @@ superuser() {
 	fi
 }
 
-nltk_data () {
-	# Store the NLTK data outside the Docker container
-	local nltk_data_dir="${DATA_DIR}/nltk"
-
-	# Download or update the snowball stemmer data
-	python3 -m nltk.downloader -d "${nltk_data_dir}" snowball_data
-
-	# Download or update the stopwords corpus
-	python3 -m nltk.downloader -d "${nltk_data_dir}" stopwords
-
-	# Download or update the punkt tokenizer data
-	python3 -m nltk.downloader -d "${nltk_data_dir}" punkt
-
-	# Set env so nltk can find the downloaded data
-	export NLTK_DATA="${nltk_data_dir}"
-
-}
-
 do_work() {
 	if [[ "${PAPERLESS_DBENGINE}" == "mariadb" ]]; then
 		wait_for_mariadb
@@ -118,8 +100,6 @@ do_work() {
 
 	migrations
 
-	nltk_data
-
 	search_index
 
 	superuser

src/documents/classifier.py

@@ -306,6 +306,12 @@ class DocumentClassifier:
         from nltk.corpus import stopwords
         from nltk.stem import SnowballStemmer
 
+        import nltk
+
+        # Not really hacky, since it isn't private and is documented, but
+        # set the search path for NLTK data to the single location it should be in
+        nltk.data.path = [settings.NLTK_DIR]
+
         if self.stemmer is None:
             self.stemmer = SnowballStemmer("english")
 

src/paperless/settings.py

@@ -84,6 +84,8 @@ THUMBNAIL_DIR = os.path.join(MEDIA_ROOT, "documents", "thumbnails")
 
 DATA_DIR = __get_path("PAPERLESS_DATA_DIR", os.path.join(BASE_DIR, "..", "data"))
 
+NLTK_DIR = os.path.join(DATA_DIR, "nltk")
+
 TRASH_DIR = os.getenv("PAPERLESS_TRASH_DIR")
 
 # Lock file for synchronizing changes to the MEDIA directory across multiple