Remove non-determinism

To enable packages using Sphinx to build reproducibly, its output
needs to be the same from one build to another.

Its output now strips memory references such as:

  <__main__.A at 0x7f68cb685710>

In addition, various generated files (objects.inv, searchindex.js,
translations) are now written with their keys in a determinstic order.

Based on a patch by Chris Lamb <lamby@debian.org>.

1 files changed, 9 insertions, 2 deletions

diff --git a/sphinx/util/inspect.py b/sphinx/util/inspect.py
index 8cbf59bde..d2d5dd0a3 100644
--- a/sphinx/util/inspect.py
+++ b/sphinx/util/inspect.py
@@ -9,6 +9,8 @@
     :license: BSD, see LICENSE for details.
 """
 
+import re
+
 # this imports the standard library inspect module without resorting to
 # relatively import this module
 inspect = __import__('inspect')
@@ -18,6 +20,8 @@ from six.moves import builtins
 
 from sphinx.util import force_decode
 
+memory_address_re = re.compile(r' at 0x[0-9a-f]{8,16}(?=>$)')
+
 
 if PY3:
     from functools import partial
@@ -123,14 +127,17 @@ def safe_getmembers(object, predicate=None, attr_getter=safe_getattr):
     return results
 
 
-def safe_repr(object):
+def object_description(object):
     """A repr() implementation that returns text safe to use in reST context."""
     try:
         s = repr(object)
     except Exception:
         raise ValueError
     if isinstance(s, binary_type):
-        return force_decode(s, None).replace('\n', ' ')
+        s = force_decode(s, None)
+    # Strip non-deterministic memory addresses such as
+    # ``<__main__.A at 0x7f68cb685710>``
+    s = memory_address_re.sub('', s)
     return s.replace('\n', ' ')