I know it's early days yet, but I've tried to create an MSI for Python 3.3. I was able to do so, but I had to make some changes to msi.py (patch attached).

This looks fine, except for the Tk changes, which should not be made.

Thanks, Martin. I've realised there are more additions due to packaging - for example there is a whole set of "wininst-X.Y[-amd64].exe" files in Lib/packaging/command. I won't commit anything just yet - I've made the changes I need to my local branch for testing.

BTW I made those Tcl/Tk changes because the instructions point to getting the files from the svn repository (using svn export), and they unpack into directories with a hyphen before the "8".

> BTW I made those Tcl/Tk changes because the instructions point to
> getting the files from the svn repository (using svn export), and
> they unpack into directories with a hyphen before the "8".

Yes, that's unfortunate. These instructions are not correct (they
are not wrong, either); they don't support Tix, but Tix support
is required for the official releases.

> I've realised there are more additions due to packaging - for example
> there is a whole set of "wininst-X.Y[-amd64].exe" files

Oh thanks, I had forgotten about msi.py.  Copying the similar section that already exists for distutils wininst executables and adding the two files added for packaging should do the trick.

> and adding the two files added for packaging should do the trick

Which two files would those be, exactly? In my branch I've changed to logic from parent.physical == "distutils" to parent.physical in ("distutils", "packaging") and this should cover the all the .exes and command_template. I haven't looked at the packaging test stuff, there may need to be some attention directed there.

Packaging has two wininst*.exe files that distutils has not.  If the code already uses a glob, it should be fine; if it uses a fixed list, then we have to add an if block.

> Packaging has two wininst*.exe files that distutils has  not.  If the code 
>already uses a glob, it should be fine; if it uses a  fixed list, then we have 
>to add an if  block.

Yes, it's using glob :-)

New changeset 3ce11ddc9ec3 by Vinay Sajip in branch 'default':
Issue #12406: Added upates for packaging's .exe files, command_template, and sysconfig.cfg.
http://hg.python.org/cpython/rev/3ce11ddc9ec3

Another problem in creating the MSI is that Lib\test\keycert.passwd.pem and Lib\test\keycert.pem both lead to a short name of 'KEYCERT.PEM', which is causing an assertion failure in msilib.Directory.make_short(). The other clashing pair is Lib\test\ssl_key.passwd.pem and Lib\test\ssl_key.pem, at least at the moment, both of which map to 'SSL_KEY.PEM'.

I also noticed that subdirectories in packaging/tests/ need to be added (e.g. fake_dists), otherwise packaging tests fail when run in an installed Python.

Adding Antoine for the short-name conflicts caused by the *.passwd.pem files.

Can't you use the normal algorithm for short names? I.e. KEYCER~1.PEM etc. Or doesn't it exist anymore?

New changeset 2fed68cf102a by Martin v. Löwis in branch 'default':
Issue #12406: prevent case where shortened name could conflict with short name.
http://hg.python.org/cpython/rev/2fed68cf102a

New changeset 75990a013d4d by Martin v. Löwis in branch 'default':
Issue #12406: Update list of files.
http://hg.python.org/cpython/rev/75990a013d4d

msi.py is now up-to-date as of 75990a013d4d. If further changes are made to the tree that require further changes to msi.py, a new issue should be opened.