Keep an application name a header cannot carry out of the User-Agent

README.md

@@ -155,6 +155,10 @@ key or value is `None` is left out, and every other key and value is sent as a
 string. An event with a blank `name` is not sent at all. The server assigns the
 timestamp. A `201` is success; anything else is logged at `DEBUG` and dropped.
 
+The `User-Agent` is `trace-client-python/<client version> (<application>)`. Any
+character of the application's name that a header cannot carry (outside printable
+Latin-1) becomes `?` there; `application` in the body is always the exact name.
+
 ## Keys
 
 A key identifies the program to the server and lets the operator revoke it;

tests/test_trace_client.py

@@ -18,7 +18,7 @@
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from unittest import mock
 
-from trace_client import MAX_TAG_LENGTH, MAX_TAGS, TraceClient, environment_opts_out
+from trace_client import MAX_TAG_LENGTH, MAX_TAGS, TraceClient, __version__, environment_opts_out
 
 _ENV_VARS = ("TRACE_USAGE_REPORTING", "DO_NOT_TRACK")
 
@@ -123,6 +123,22 @@ def test_non_ascii_names_and_tags_arrive_intact_as_utf8_json(self):
                           "tags": {"ville": "Zürich", "emoji": "\U0001F3B2", "version": "1.2.3"}},
                          json.loads(request["body"]), "text the program reports is what the server reads")
 
+    def test_an_application_name_a_header_cannot_carry_still_delivers_its_events(self):
+        # The name goes in the User-Agent as well as the body, and a header
+        # refuses most of Unicode and any line break.
+        for application, shown in (("ゲーム", "???"), ("dice\U0001F3B2", "dice?"), ("two\nlines", "two?lines"),
+                                   ("Zürich", "Zürich")):
+            self.capture.requests.clear()
+            client = TraceClient(self.base_url, application, "1.2.3", key="k")
+            client.report("startup")
+            client.close()
+            self.assertEqual(1, len(self.capture.requests), "%r lost its event: %s"
+                             % (application, [r.getMessage() for r in self.log]))
+            request = self.capture.requests[0]
+            self.assertEqual(application, json.loads(request["body"])["application"],
+                             "the event keeps the exact name; only the header is approximate")
+            self.assertEqual("trace-client-python/%s (%s)" % (__version__, shown), request["user_agent"])
+
     def test_report_carries_value_and_tags_when_given(self):
         client = TraceClient(self.base_url, "MyGame", "1.2.3", key="k")
         client.report("world-load", 2.5, {"seed": "42", "size": 'the "big" one'})

trace_client/trace_client.py

@@ -53,6 +53,9 @@
 # What install_id_from_file accepts as an ID on a line of its file.
 _INSTALL_ID_LINE = re.compile(r"^[A-Za-z0-9_.-]{1,%d}$" % MAX_TAG_LENGTH)
 
+# What the User-Agent header cannot carry: everything but printable Latin-1.
+_NOT_IN_A_HEADER = re.compile(r"[^\x20-\x7e\xa0-\xff]")
+
 
 def environment_opts_out(environ: Optional[Mapping[str, str]] = None) -> bool:
     """Whether the environment asks for usage reporting to be off, via
@@ -308,7 +311,7 @@ def _send(self, body: bytes) -> None:
                 headers={
                     "Content-Type": "application/json; charset=utf-8",
                     "Authorization": "Bearer " + self._key,
-                    "User-Agent": "trace-client-python/%s (%s)" % (__version__, self._application),
+                    "User-Agent": _user_agent(self._application),
                 })
             with urllib.request.urlopen(request, timeout=self.TIMEOUT_SECONDS) as response:
                 status = response.status
@@ -326,6 +329,14 @@ def _send(self, body: bytes) -> None:
             _LOG.debug("[trace] trace server answered %s for %s", status, body)
 
 
+def _user_agent(application: str) -> str:
+    """The ``User-Agent`` for ``application``. A header carries printable
+    Latin-1 and nothing else, and one it cannot carry stops the request, so
+    any other character becomes ``?`` here. The event itself keeps the exact
+    name: what a program is called must not cost it every report."""
+    return "trace-client-python/%s (%s)" % (__version__, _NOT_IN_A_HEADER.sub("?", application))
+
+
 def _with_version(tags: Optional[Mapping[str, str]], version: str) -> Dict[str, str]:
     """The event's own tags plus ``version``, unless the event already carries
     one. A copy; the caller's mapping is never modified."""