[spambayes-dev] Open relay, and new release?

[Mark Hammond]

> I'm not sure how closely you're monitoring the bug reports at
> the moment, so
> in case you're not:

I'm not :)

> Someone that had this problem searched for all instances of
> the pythoncom
> dll, found two, and replaced the system one with the
> spambayes one, and the
> problem went away (although who knows what other trouble that
> will cause!).
> Another user, reading the comment, did the same thing, and
> his problem went
> away, too.

I have discovered the problem.  ActivePython is still registering the
'pythoncom' and 'pywintypes' module in the registry.  This is no longer
necessary, for any version of Python.  Indeed, it is what is causing the
problem - our binary still *does* look in the registry, as python23.dll
itself does.

The capability to register modules in the registry was a dumb one that I
must take full credit for <wink>.  Trent - would it be possible for you to
modify the ActivePython installer to not write these module values, now
pywin32 no longer needs them?

The best thing I can see to do is for us to ship a slightly modified
python23.dll.  All that needs modification is a string resource in the DLL -
ie, no code.  This will change where Python looks for its configuration
information in the registry.  By changing this value, we can ensure that we
can't possibly conflict with any existing installations.

At this stage, the only way I can change the resource is via MSVC.  With a
little work, it will be possible to have py2exe write a custom value for
this string into the DLL - it can already write resources, but strings need
a little work.  Thomas - if you are listening, do you have any thoughts on
this?  Unfortunately, we really can't simply ask people to upgrade
ActivePython, as I believe even the latest writes this registry value.
Further, an upgrade of an old ActivePython may leave it in place (recent
pywin32 installers actually remove the values!)

And finally, I'm off for a week holiday.  I tried to get a build out
(including a hand-modified string resource), but for some reason it is
failing to work for me.  I'm out of time to debug this, so I'm afraid it
will have to wait.

Thanks,

Mark.