[Zope-Perl] security

[Chris McDonough]

Gisle,

When a Perl Method returns a reference to a hash (or list), a perl ref
object is returned to DTML or Python.  This is really neat.

There's one problem I can see, however:  Zope's security machinery inside
DTML and Python Scripts expects instances to define a "__roles__" attribute
so they may be accessed in TTW code.  Usually, I would do something like
(from PerlMethod __init__.py's __call__ method):

 r = apply(perl.safecall, (root, mask, args))
 roles = self.aq_acquire('__roles__') # grabs __roles__ of PerlM's
container.
 setattr(r, '__roles__', roles)
 return r

.. but I can't setattr on the perlref instance, so this doesn't work.

Without jiggering of the security machinery or a way to set __roles__ on
perlref instances, doing (for example) in DTML:

<dtml-in>
  <dtml-with "perlmwhichreturnshashref()">
     <dtml-in keys>
        <dtml-var sequence-item>
     </dtml-in>
  </dtml-with>
</dtml-in>

... doesn't work (the Zope security policy raises unauthorized at the point
in which you attempt to access keys).

I *think* this can be solved in a couple of ways (each radically different):

- Do type conversion on hashrefs and arrayrefs to Python native datatypes
(dict, list) in a return to Python if the hash or array keys/values are all
string or number types.  This is potentially very expensive, and a little
DWIMish.

- Allow perlref instances in Python to have a __dict__ and let us assign to
it from unrestricted code (so we can give it a __roles__).

- Jigger the security machinery to do more DWIM.

Thoughts?