subject.location in dublin core
[librarian.git] / tests / test_html.py
index 4fbbaad..a0de630 100644 (file)
@@ -3,44 +3,39 @@
 # This file is part of Librarian, licensed under GNU Affero GPLv3 or later.
 # Copyright © Fundacja Nowoczesna Polska. See NOTICE for more information.
 #
-from librarian import html, NoDublinCore
+from librarian import NoDublinCore
+from librarian.parser import WLDocument
 from nose.tools import *
-from utils import get_fixture, remove_output_file
+from utils import get_fixture
 
-def teardown_transform():
-    remove_output_file('text', 'asnyk_miedzy_nami.html')
 
-
-@with_setup(None, teardown_transform)
 def test_transform():
-    output_file_path = get_fixture('text', 'asnyk_miedzy_nami.html')
     expected_output_file_path = get_fixture('text', 'asnyk_miedzy_nami_expected.html')
 
-    html.transform(
-        get_fixture('text', 'asnyk_miedzy_nami.xml'),
-        output_file_path,
-    )
+    html = WLDocument.from_file(
+            get_fixture('text', 'miedzy-nami-nic-nie-bylo.xml')
+        ).as_html().get_string()
 
-    assert_equal(file(output_file_path).read(), file(expected_output_file_path).read())
+    assert_equal(html, file(expected_output_file_path).read())
 
 
-@with_setup(None, teardown_transform)
 @raises(NoDublinCore)
 def test_no_dublincore():
-    html.transform(
-        get_fixture('text', 'asnyk_miedzy_nami_nodc.xml'),
-        get_fixture('text', 'asnyk_miedzy_nami_nodc.html'),
-    )
+    WLDocument.from_file(
+            get_fixture('text', 'asnyk_miedzy_nami_nodc.xml')
+        ).as_html()
 
 
-@with_setup(None, teardown_transform)
 def test_passing_parse_dublincore_to_transform():
     """Passing parse_dublincore=False to transform omits DublinCore parsing."""
-    html.transform(
-        get_fixture('text', 'asnyk_miedzy_nami_nodc.xml'),
-        get_fixture('text', 'asnyk_miedzy_nami.html'),
-        parse_dublincore=False,
-    )
+    WLDocument.from_file(
+            get_fixture('text', 'asnyk_miedzy_nami_nodc.xml'),
+            parse_dublincore=False,
+        ).as_html()
+
 
 def test_empty():
-    assert html.transform('<utwor />', is_file=False, parse_dublincore=False).find('empty')
+    assert not WLDocument.from_string(
+            '<utwor />',
+            parse_dublincore=False,
+        ).as_html()