Change pecan to more gracefully handle a few odd request encoding edge cases.

Webob raises UnicodeDecodeErrors during request argument parsing in two
situations:
    * HTTP POST requests composed of non-Unicode data (only affects Webob in
      Python2)
    * URL paths that contain invalid percent-encoded characters, e.g.,
      /some/path/%AA

Pecan should detect these types of decoding failures when `webob.Request`
attributes are accessed, log the original exception, and coerce the HTTP 500
into a more accurate HTTP 400.

Fixes bug: 1408102
Fixes bug: 1407749

Change-Id: I734efd36230b6742805bcfd801dc0de2489ef92b

3 files changed, 48 insertions, 3 deletions

diff --git a/pecan/core.py b/pecan/core.py
index 173210b..e74a8b1 100644
--- a/pecan/core.py
+++ b/pecan/core.py
@@ -48,7 +48,13 @@ class RoutingState(object):
 
 
 class Request(WebObRequest):
-    pass
+
+    def __getattribute__(self, name):
+        try:
+            return WebObRequest.__getattribute__(self, name)
+        except UnicodeDecodeError as e:
+            logger.exception(e)
+            abort(400)
 
 
 class Response(WebObResponse):
diff --git a/pecan/tests/test_base.py b/pecan/tests/test_base.py
index cca494c..1369837 100644
--- a/pecan/tests/test_base.py
+++ b/pecan/tests/test_base.py
@@ -146,6 +146,23 @@ class TestAppIterFile(PecanTestCase):
         assert len(r.body) == 0
 
 
+class TestInvalidURLEncoding(PecanTestCase):
+
+    @property
+    def app_(self):
+        class RootController(object):
+
+            @expose()
+            def _route(self, args, request):
+                assert request.path
+
+        return TestApp(Pecan(RootController()))
+
+    def test_rest_with_non_utf_8_body(self):
+        r = self.app_.get('/%aa/', expect_errors=True)
+        assert r.status_int == 400
+
+
 class TestIndexRouting(PecanTestCase):
 
     @property
diff --git a/pecan/tests/test_rest.py b/pecan/tests/test_rest.py
index d839084..391b7d0 100644
--- a/pecan/tests/test_rest.py
+++ b/pecan/tests/test_rest.py
@@ -1,11 +1,12 @@
-from webtest import TestApp
+import struct
 import warnings
 try:
     from simplejson import dumps, loads
 except:
     from json import dumps, loads  # noqa
 
-from six import b as b_
+from six import b as b_, PY3
+from webtest import TestApp
 
 from pecan import abort, expose, make_app, response, redirect
 from pecan.rest import RestController
@@ -1359,6 +1360,27 @@ class TestRestController(PecanTestCase):
         assert r.status_int == 200
         assert r.body == b_("DEFAULT missing")
 
+    def test_rest_with_non_utf_8_body(self):
+        if PY3:
+            # webob+PY3 doesn't suffer from this bug; the POST parsing in PY3
+            # seems to more gracefully detect the bytestring
+            return
+
+        class FooController(RestController):
+
+            @expose()
+            def post(self):
+                return "POST"
+
+        class RootController(RestController):
+            foo = FooController()
+
+        app = TestApp(make_app(RootController()))
+
+        data = struct.pack('255h', *range(0, 255))
+        r = app.post('/foo/', data, expect_errors=True)
+        assert r.status_int == 400
+
     def test_dynamic_rest_lookup(self):
         class BarController(RestController):
             @expose()