In dictobject.c the structures from dictobject.h are typedeffed to:
typedef PyDictEntry dictentry;
typedef PyDictObject dictobject;

However there are still a few locations in that file where the
PyDictEntry and PyDictObject types are used directly. IMHO these should
be replaced by  dictentry resp. dictobject

Attached is a patch for that.

I think it would be better to get rid of the typedefs and use the public
Py* names everywhere?

Guido's suggestion to change all entries to PyDictEntry resp.
PyDictObject would work as well and declutter the code in a better way.

The only advantage of the typedefs that I see (and briefly used) it that
it is easy to have structures local to dictobject.c that hold some more
information, without having to touch dictobject.h (and trigger a more
general recompile. For the few that need that, they can globaly replace
PyDict{Entry,Object} when they do.

While I understand the argument for faster recompiles, dictobject.(c|h)
do not change that often, and thus faster recompiles are not critical. 
I am with Guido and would rather see the module moved over to public names.

Setting the priority to low as this is not critical in any way, although
I am all for making code more readable and thus will review any patch
that Anthon comes up with that uses the public names.

I went ahead and had Vim do a global search-and-replace on the names and
ditched the typedefs; the patch is uploaded.  I have not yet run the
test suite, though, which is why I have not committed it.  But if it
passes I will just check it in.

Applied my patch in r58399.  Thanks for noticing the inconsistency, anthon!