Don't decode YAML responses before parsing them in get_artifact

Taskcluster sends those without a content-type which means that requests
assumes they're latin1 text, which they might not be as taskgraph uploads
its YAMLs encoded as UTF-8. While decoding UTF-8 as latin1 works most of
the time, if the commit message passed in a parameters.yml contains a
special character, then it will return a garbled mess which the yaml
library will reject.

By passing the raw bytes instead of decoding them, we sidestep the
problem entirely, delegating the decoding to the YAML library which
supports both utf-8 and utf-16s.

From the documentation for `yaml.load`

> A byte string or a file must be encoded with utf-8, utf-16-be or
> utf-16-le encoding. yaml.load detects the encoding by checking the BOM
> (byte order mark) sequence at the beginning of the string/file. If no
> BOM is present, the utf-8 encoding is assumed.

Author: Eijebong

src/taskgraph/util/taskcluster.py

@@ -143,7 +143,7 @@ def _handle_artifact(path, response):
     if path.endswith(".json"):
         return response.json()
     if path.endswith(".yml"):
-        return yaml.load_stream(response.text)
+        return yaml.load_stream(response.content)
     response.raw.read = functools.partial(response.raw.read, decode_content=True)
     return response.raw
 

test/test_util_taskcluster.py

@@ -154,6 +154,21 @@ def test_get_artifact(responses, root_url):
     )
     assert tc.get_artifact(tid, "artifact.yml") == {"foo": "bar"}
 
+    responses.add(
+        responses.GET,
+        f"{root_url}/api/queue/v1/task/{tid}/artifacts/artifact.yml",
+        body=b"foo: \xe2\x81\x83",
+    )
+    assert tc.get_artifact(tid, "artifact.yml") == {"foo": b"\xe2\x81\x83".decode()}
+
+    responses.add(
+        responses.GET,
+        f"{root_url}/api/queue/v1/task/{tid}/artifacts/artifact.yml",
+        body=b"foo: \xe2\x81\x83".decode().encode("utf-16"),
+        headers={"Content-Type": "text/yaml; charset=utf-16"},
+    )
+    assert tc.get_artifact(tid, "artifact.yml") == {"foo": b"\xe2\x81\x83".decode()}
+
 
 def test_list_artifact(responses, root_url):
     tid = 123