Add a representer for AnsibleUnsafeBytes (#62598)

* Add a representer for AnsibleUnsafeBytes

* changelog

* Add unit tests

Remove native string test until we have time to evaluate how this the function should work
Add non-ASCII characters to test cases

* Compare to the string on Python 2

Add a comment in the test about this behavior

changelogs/fragments/62598-AnsibleDumper-representer.yaml

@@ -0,0 +1,2 @@
+bugfixes:
+  - AnsibleDumper - Add a representer for AnsibleUnsafeBytes (https://github.com/ansible/ansible/issues/62562).

lib/ansible/parsing/yaml/dumper.py

@@ -23,7 +23,7 @@ import yaml
 
 from ansible.module_utils.six import PY3
 from ansible.parsing.yaml.objects import AnsibleUnicode, AnsibleSequence, AnsibleMapping, AnsibleVaultEncryptedUnicode
-from ansible.utils.unsafe_proxy import AnsibleUnsafeText
+from ansible.utils.unsafe_proxy import AnsibleUnsafeText, AnsibleUnsafeBytes
 from ansible.vars.hostvars import HostVars, HostVarsVars
 
 
@@ -46,8 +46,10 @@ def represent_vault_encrypted_unicode(self, data):
 
 if PY3:
     represent_unicode = yaml.representer.SafeRepresenter.represent_str
+    represent_binary = yaml.representer.SafeRepresenter.represent_binary
 else:
     represent_unicode = yaml.representer.SafeRepresenter.represent_unicode
+    represent_binary = yaml.representer.SafeRepresenter.represent_str
 
 AnsibleDumper.add_representer(
     AnsibleUnicode,
@@ -59,6 +61,11 @@ AnsibleDumper.add_representer(
     represent_unicode,
 )
 
+AnsibleDumper.add_representer(
+    AnsibleUnsafeBytes,
+    represent_binary,
+)
+
 AnsibleDumper.add_representer(
     HostVars,
     represent_hostvars,

test/units/parsing/yaml/test_dumper.py

@@ -24,6 +24,8 @@ from units.compat import unittest
 from ansible.parsing import vault
 from ansible.parsing.yaml import dumper, objects
 from ansible.parsing.yaml.loader import AnsibleLoader
+from ansible.module_utils.six import PY2
+from ansible.utils.unsafe_proxy import AnsibleUnsafeText, AnsibleUnsafeBytes
 
 from units.mock.yaml_helper import YamlTestUtils
 from units.mock.vault_helper import TextVaultSecret
@@ -47,7 +49,7 @@ class TestAnsibleDumper(unittest.TestCase, YamlTestUtils):
     def _loader(self, stream):
         return AnsibleLoader(stream, vault_secrets=self.vault.secrets)
 
-    def test(self):
+    def test_ansible_vault_encrypted_unicode(self):
         plaintext = 'This is a string we are going to encrypt.'
         avu = objects.AnsibleVaultEncryptedUnicode.from_plaintext(plaintext, vault=self.vault,
                                                                   secret=vault.match_secrets(self.vault_secrets, ['vault_secret'])[0][1])
@@ -59,3 +61,43 @@ class TestAnsibleDumper(unittest.TestCase, YamlTestUtils):
         data_from_yaml = loader.get_single_data()
 
         self.assertEqual(plaintext, data_from_yaml.data)
+
+    def test_bytes(self):
+        b_text = u'tréma'.encode('utf-8')
+        unsafe_object = AnsibleUnsafeBytes(b_text)
+        yaml_out = self._dump_string(unsafe_object, dumper=self.dumper)
+
+        stream = self._build_stream(yaml_out)
+        loader = self._loader(stream)
+
+        data_from_yaml = loader.get_single_data()
+
+        result = b_text
+        if PY2:
+            # https://pyyaml.org/wiki/PyYAMLDocumentation#string-conversion-python-2-only
+            # pyyaml on Python 2 can return either unicode or bytes when given byte strings.
+            # We normalize that to always return unicode on Python2 as that's right most of the
+            # time.  However, this means byte strings can round trip through yaml on Python3 but
+            # not on Python2.  To make this code work the same on Python2 and Python3 (we want
+            # the Python3 behaviour) we need to change the methods in Ansible to:
+            # (1) Let byte strings pass through yaml without being converted on Python2
+            # (2) Convert byte strings to text strings before being given to pyyaml  (Without this,
+            #       strings would end up as byte strings most of the time which would mostly be wrong)
+            # In practice, we mostly read bytes in from files and then pass that to pyyaml, for which
+            # the present behavior is correct.
+            # This is a workaround for the current behavior.
+            result = u'tr\xe9ma'
+
+        self.assertEqual(result, data_from_yaml)
+
+    def test_unicode(self):
+        u_text = u'nöel'
+        unsafe_object = AnsibleUnsafeText(u_text)
+        yaml_out = self._dump_string(unsafe_object, dumper=self.dumper)
+
+        stream = self._build_stream(yaml_out)
+        loader = self._loader(stream)
+
+        data_from_yaml = loader.get_single_data()
+
+        self.assertEqual(u_text, data_from_yaml)