Backend: remove invalid/deprecated ISO639-3 language codes

The allowed-languages list had three bad codes: `_tw` (invented; the official
code is `nan`), `ajp` (deprecated, merged into `apc` per ISO 639-3 CR 2022-006),
and `smi` (a part-2 group code, not an individual language). Remove them from
languages.json and add migration 0170 to repoint existing user language
abilities (`_tw`->`nan`, `ajp`->`apc`, drop `smi`), deduping to respect the
unique (user_id, language_code) constraint. Adds a regression test that loads
the real languages.json and asserts the codes are gone.

Fixes #9172

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: kevinortiz43

app/backend/resources/languages.json

@@ -1,8 +1,4 @@
 [
-  {
-    "code": "_tw",
-    "name": "Taiwanese"
-  },
   {
     "code": "aao",
     "name": "Arabic (Algerian)"
@@ -55,10 +51,6 @@
     "code": "afr",
     "name": "Afrikaans"
   },
-  {
-    "code": "ajp",
-    "name": "Arabic (Levantine South)"
-  },
   {
     "code": "aka",
     "name": "Akan"
@@ -947,10 +939,6 @@
     "code": "slv",
     "name": "Slovenian"
   },
-  {
-    "code": "smi",
-    "name": "Sámi"
-  },
   {
     "code": "smo",
     "name": "Samoan"

app/backend/src/couchers/migrations/versions/0170_fix_invalid_language_codes.py

@@ -0,0 +1,45 @@
+"""Fix invalid/deprecated ISO 639-3 language codes
+
+Revision ID: 0170
+Revises: 0169
+Create Date: 2026-06-29 00:00:00.000000
+
+"""
+
+from alembic import op
+from sqlalchemy.orm.session import Session
+
+from couchers.resources import copy_resources_to_database
+
+# revision identifiers, used by Alembic.
+revision = "0170"
+down_revision = "0169"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    session = Session(bind=op.get_bind())
+    # Reload languages table (removes _tw, ajp, smi from the static table)
+    copy_resources_to_database(session)
+    # _tw ("Taiwanese") is an invented code; official equivalent is nan ("Chinese (Southern Min)")
+    # Dedupe first: drop _tw for users who already have nan
+    op.execute(
+        "DELETE FROM language_abilities WHERE language_code='_tw' AND user_id IN "
+        "(SELECT user_id FROM language_abilities WHERE language_code='nan');"
+    )
+    op.execute("UPDATE language_abilities SET language_code='nan' WHERE language_code='_tw';")
+    # ajp ("Arabic (Levantine South)") is deprecated; ISO merged it into apc ("Arabic (Levantine North)")
+    # Dedupe first: drop ajp for users who already have apc
+    op.execute(
+        "DELETE FROM language_abilities WHERE language_code='ajp' AND user_id IN "
+        "(SELECT user_id FROM language_abilities WHERE language_code='apc');"
+    )
+    op.execute("UPDATE language_abilities SET language_code='apc' WHERE language_code='ajp';")
+    # smi ("Sámi") is an ISO 639-2 group code with no individual equivalent; drop it
+    op.execute("DELETE FROM language_abilities WHERE language_code='smi';")
+    session.commit()
+
+
+def downgrade() -> None:
+    pass

app/backend/src/tests/test_resources.py

@@ -1,6 +1,10 @@
 import pytest
 from google.protobuf import empty_pb2
+from sqlalchemy import select
 
+from couchers.db import session_scope
+from couchers.models import Language
+from couchers.resources import copy_resources_to_database
 from tests.fixtures.sessions import resources_session
 
 
@@ -57,6 +61,21 @@ def test_GetLanguages(db):
         assert ("swe", "Swedish") not in languages_list
 
 
+def test_languages_resource_excludes_invalid_codes(db):
+    # Load the real languages.json (not the hardcoded testing fixture) so a future bad code is caught.
+    # Mirrors test_add_dummy_data's pattern of calling copy_resources_to_database directly.
+    with session_scope() as session:
+        copy_resources_to_database(session)
+        codes = set(session.execute(select(Language.code)).scalars().all())
+    # Codes removed in #9172: invalid/deprecated ISO 639-3 entries
+    assert "_tw" not in codes  # invented code; use nan (Chinese Southern Min) instead
+    assert "ajp" not in codes  # deprecated; ISO merged into apc (Arabic Levantine North)
+    assert "smi" not in codes  # ISO 639-2 group code with no individual equivalent
+    # Remap targets must remain present
+    assert "nan" in codes
+    assert "apc" in codes
+
+
 def test_GetBadges(db):
     with resources_session() as api:
         badges = api.GetBadges(empty_pb2.Empty()).badges