Project import generated by Copybara.

GitOrigin-RevId: d9e9e3fb4e31372ec1fb43b178994ca78fa8fe70
diff --git a/framework/test/template_helpers_test.py b/framework/test/template_helpers_test.py
new file mode 100644
index 0000000..85296fa
--- /dev/null
+++ b/framework/test/template_helpers_test.py
@@ -0,0 +1,216 @@
+# Copyright 2016 The Chromium Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style
+# license that can be found in the LICENSE file or at
+# https://developers.google.com/open-source/licenses/bsd
+
+"""Unit tests for template_helpers module."""
+
+from __future__ import division
+from __future__ import print_function
+from __future__ import absolute_import
+
+import unittest
+
+from framework import pbproxy_test_pb2
+from framework import template_helpers
+
+
+class HelpersUnitTest(unittest.TestCase):
+
+  def testDictionaryProxy(self):
+
+    # basic in 'n out test
+    item = template_helpers.EZTItem(label='foo', group_name='bar')
+
+    self.assertEqual('foo', item.label)
+    self.assertEqual('bar', item.group_name)
+
+    # be sure the __str__ returns the fields
+    self.assertEqual(
+        "EZTItem({'group_name': 'bar', 'label': 'foo'})", str(item))
+
+  def testPBProxy(self):
+    """Checks that PBProxy wraps protobuf objects as expected."""
+    # check that protobuf fields are accessible in ".attribute" form
+    pbe = pbproxy_test_pb2.PBProxyExample()
+    pbe.nickname = 'foo'
+    pbe.invited = False
+    pbep = template_helpers.PBProxy(pbe)
+    self.assertEqual(pbep.nickname, 'foo')
+    # _bool suffix converts protobuf field 'bar' to None (EZT boolean false)
+    self.assertEqual(pbep.invited_bool, None)
+
+    # check that a new field can be added to the PBProxy
+    pbep.baz = 'bif'
+    self.assertEqual(pbep.baz, 'bif')
+
+    # check that a PBProxy-local field can hide a protobuf field
+    pbep.nickname = 'local foo'
+    self.assertEqual(pbep.nickname, 'local foo')
+
+    # check that a nested protobuf is recursively wrapped with a PBProxy
+    pbn = pbproxy_test_pb2.PBProxyNested()
+    pbn.nested = pbproxy_test_pb2.PBProxyExample()
+    pbn.nested.nickname = 'bar'
+    pbn.nested.invited = True
+    pbnp = template_helpers.PBProxy(pbn)
+    self.assertEqual(pbnp.nested.nickname, 'bar')
+    # _bool suffix converts protobuf field 'bar' to 'yes' (EZT boolean true)
+    self.assertEqual(pbnp.nested.invited_bool, 'yes')
+
+    # check that 'repeated' lists of items produce a list of strings
+    pbn.multiple_strings.append('1')
+    pbn.multiple_strings.append('2')
+    self.assertEqual(pbnp.multiple_strings, ['1', '2'])
+
+    # check that 'repeated' messages produce lists of PBProxy instances
+    pbe1 = pbproxy_test_pb2.PBProxyExample()
+    pbn.multiple_pbes.append(pbe1)
+    pbe1.nickname = '1'
+    pbe1.invited = True
+    pbe2 = pbproxy_test_pb2.PBProxyExample()
+    pbn.multiple_pbes.append(pbe2)
+    pbe2.nickname = '2'
+    pbe2.invited = False
+    self.assertEqual(pbnp.multiple_pbes[0].nickname, '1')
+    self.assertEqual(pbnp.multiple_pbes[0].invited_bool, 'yes')
+    self.assertEqual(pbnp.multiple_pbes[1].nickname, '2')
+    self.assertEqual(pbnp.multiple_pbes[1].invited_bool, None)
+
+  def testFitTextMethods(self):
+    """Tests both FitUnsafeText with an eye on i18n."""
+    # pylint: disable=anomalous-unicode-escape-in-string
+    test_data = (
+        u'This is a short string.',
+
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. '
+        u'This is a much longer string. ',
+
+        # This is a short escaped i18n string
+        '\xd5\xa1\xd5\xba\xd5\xa1\xd5\xaf\xd5\xab'.decode('utf-8'),
+
+        # This is a longer i18n string
+        '\xd5\xa1\xd5\xba\xd5\xa1\xd5\xaf\xd5\xab '
+        '\xe6\x88\x91\xe8\x83\xbd\xe5\x90\x9e '
+        '\xd5\xa1\xd5\xba\xd5\xa1\xd5\xaf\xd5\xab '
+        '\xe6\x88\x91\xe8\x83\xbd\xe5\x90\x9e '
+        '\xd5\xa1\xd5\xba\xd5\xa1\xd5\xaf\xd5\xab '
+        '\xe6\x88\x91\xe8\x83\xbd\xe5\x90\x9e '
+        '\xd5\xa1\xd5\xba\xd5\xa1\xd5\xaf\xd5\xab '
+        '\xe6\x88\x91\xe8\x83\xbd\xe5\x90\x9e '.decode('utf-8'),
+
+        # This is a longer i18n string that was causing trouble.
+        '\u041d\u0430 \u0431\u0435\u0440\u0435\u0433\u0443'
+        ' \u043f\u0443\u0441\u0442\u044b\u043d\u043d\u044b\u0445'
+        ' \u0432\u043e\u043b\u043d \u0421\u0442\u043e\u044f\u043b'
+        ' \u043e\u043d, \u0434\u0443\u043c'
+        ' \u0432\u0435\u043b\u0438\u043a\u0438\u0445'
+        ' \u043f\u043e\u043b\u043d, \u0418'
+        ' \u0432\u0434\u0430\u043b\u044c'
+        ' \u0433\u043b\u044f\u0434\u0435\u043b.'
+        ' \u041f\u0440\u0435\u0434 \u043d\u0438\u043c'
+        ' \u0448\u0438\u0440\u043e\u043a\u043e'
+        ' \u0420\u0435\u043a\u0430'
+        ' \u043d\u0435\u0441\u043b\u0430\u0441\u044f;'
+        ' \u0431\u0435\u0434\u043d\u044b\u0439'
+        ' \u0447\u0451\u043b\u043d \u041f\u043e'
+        ' \u043d\u0435\u0439'
+        ' \u0441\u0442\u0440\u0435\u043c\u0438\u043b\u0441\u044f'
+        ' \u043e\u0434\u0438\u043d\u043e\u043a\u043e.'
+        ' \u041f\u043e \u043c\u0448\u0438\u0441\u0442\u044b\u043c,'
+        ' \u0442\u043e\u043f\u043a\u0438\u043c'
+        ' \u0431\u0435\u0440\u0435\u0433\u0430\u043c'
+        ' \u0427\u0435\u0440\u043d\u0435\u043b\u0438'
+        ' \u0438\u0437\u0431\u044b \u0437\u0434\u0435\u0441\u044c'
+        ' \u0438 \u0442\u0430\u043c, \u041f\u0440\u0438\u044e\u0442'
+        ' \u0443\u0431\u043e\u0433\u043e\u0433\u043e'
+        ' \u0447\u0443\u0445\u043e\u043d\u0446\u0430;'
+        ' \u0418 \u043b\u0435\u0441,'
+        ' \u043d\u0435\u0432\u0435\u0434\u043e\u043c\u044b\u0439'
+        ' \u043b\u0443\u0447\u0430\u043c \u0412'
+        ' \u0442\u0443\u043c\u0430\u043d\u0435'
+        ' \u0441\u043f\u0440\u044f\u0442\u0430\u043d\u043d\u043e'
+        '\u0433\u043e \u0441\u043e\u043b\u043d\u0446\u0430,'
+        ' \u041a\u0440\u0443\u0433\u043e\u043c'
+        ' \u0448\u0443\u043c\u0435\u043b.'.decode('utf-8'))
+
+    for unicode_s in test_data:
+      # Get the length in characters, not bytes.
+      length = len(unicode_s)
+
+      # Test the FitUnsafeText method at the length boundary.
+      fitted_unsafe_text = template_helpers.FitUnsafeText(unicode_s, length)
+      self.assertEqual(fitted_unsafe_text, unicode_s)
+
+      # Set some values that test FitString well.
+      available_space = length // 2
+      max_trailing = length // 4
+      # Break the string at various places - symmetric range around 0
+      for i in range(1-max_trailing, max_trailing):
+        # Test the FitUnsafeText method.
+        fitted_unsafe_text = template_helpers.FitUnsafeText(
+            unicode_s, available_space - i)
+        self.assertEqual(fitted_unsafe_text[:available_space - i],
+                         unicode_s[:available_space - i])
+
+      # Test a string that is already unicode
+      u_string = u'This is already unicode'
+      fitted_unsafe_text = template_helpers.FitUnsafeText(u_string, 100)
+      self.assertEqual(u_string, fitted_unsafe_text)
+
+      # Test a string that is already unicode, and has non-ascii in it.
+      u_string = u'This is already unicode este\\u0301tico'
+      fitted_unsafe_text = template_helpers.FitUnsafeText(u_string, 100)
+      self.assertEqual(u_string, fitted_unsafe_text)
+
+  def testEZTError(self):
+    errors = template_helpers.EZTError()
+    self.assertFalse(errors.AnyErrors())
+
+    errors.error_a = 'A'
+    self.assertTrue(errors.AnyErrors())
+    self.assertEqual('A', errors.error_a)
+
+    errors.SetError('error_b', 'B')
+    self.assertTrue(errors.AnyErrors())
+    self.assertEqual('A', errors.error_a)
+    self.assertEqual('B', errors.error_b)
+
+  def testBytesKbOrMb(self):
+    self.assertEqual('1023 bytes', template_helpers.BytesKbOrMb(1023))
+    self.assertEqual('1.0 KB', template_helpers.BytesKbOrMb(1024))
+    self.assertEqual('1023 KB', template_helpers.BytesKbOrMb(1024 * 1023))
+    self.assertEqual('1.0 MB', template_helpers.BytesKbOrMb(1024 * 1024))
+    self.assertEqual('98.0 MB', template_helpers.BytesKbOrMb(98 * 1024 * 1024))
+    self.assertEqual('99 MB', template_helpers.BytesKbOrMb(99 * 1024 * 1024))
+
+
+class TextRunTest(unittest.TestCase):
+
+  def testLink(self):
+    run = template_helpers.TextRun(
+        'content', tag='a', href='http://example.com')
+    expected = '<a href="http://example.com">content</a>'
+    self.assertEqual(expected, run.FormatForHTMLEmail())
+
+    run = template_helpers.TextRun(
+      'con<tent>', tag='a', href='http://exa"mple.com')
+    expected = '<a href="http://exa&quot;mple.com">con&lt;tent&gt;</a>'
+    self.assertEqual(expected, run.FormatForHTMLEmail())
+
+  def testText(self):
+    run = template_helpers.TextRun('content')
+    expected = 'content'
+    self.assertEqual(expected, run.FormatForHTMLEmail())
+
+    run = template_helpers.TextRun('con<tent>')
+    expected = 'con&lt;tent&gt;'
+    self.assertEqual(expected, run.FormatForHTMLEmail())