Tuple unpacking to local variables should be allowed

[tseaver]

E.g., from the Products.PythonScripts fibonacci test:

l = []
a, b = 0, 1
while b < 100000000:
    l.append(b)
    a, b = b, a+b  # fails with error
return l

The error is:

> /home/tseaver/.buildout/eggs/cp27m/RestrictedPython-4.0a1-py2.7.egg/RestrictedPython/tests/verify.py(45)verify()
-> verify(ob)
(Pdb) d
> /home/tseaver/.buildout/eggs/cp27m/RestrictedPython-4.0a1-py2.7.egg/RestrictedPython/tests/verify.py(42)verify()
-> verifycode(code)
(Pdb) d
> /home/tseaver/.buildout/eggs/cp27m/RestrictedPython-4.0a1-py2.7.egg/RestrictedPython/tests/verify.py(50)verifycode()
-> _verifycode(code)
(Pdb) d
> /home/tseaver/.buildout/eggs/cp27m/RestrictedPython-4.0a1-py2.7.egg/RestrictedPython/tests/verify.py(110)_verifycode()
-> (code.co_filename, line))
(Pdb) l
105  	            if not (window[0].opname == "CALL_FUNCTION" and
106  	                    window[1].opname == "ROT_TWO" and
107  	                    window[2].opname == "LOAD_GLOBAL" and
108  	                    window[2].arg == "_getiter_"):
109  	                raise ValueError("unguarded unpack sequence at %s:%d" %
110  ->	                                 (code.co_filename, line))
111  	
112  	        # should check CALL_FUNCTION_{VAR,KW,VAR_KW} but that would
113  	        # require a potentially unlimited history.  need to refactor
114  	        # the "window" before I can do that.
115  	

[icemac]

@tseaver Do you have a link to the code of the failing test?
It still uses the verify function which does some checks on byte-code level. These checks are no longer correct for tuple unpacking as it is now implemented using _unpack_sequence_ with a different byte code than before.

We dropped using verify on other places as we were not able to adapt it. See also the PendingDeprecationWarning when using this function.

[stephan-hof]

As far as I can see the following is causing problems:
https://github.com/zopefoundation/Products.PythonScripts/blob/master/src/Products/PythonScripts/tests/testPythonScript.py#L62

So the tests of Products.PythonScripts use this verify on each script before they run it. Currently there is no verify available so the only suggestion I have is removing this line.

Let my try to explain why the verify was dropped and not ported.
The new implementation of RestrictedPython is not using bytecode transformations at all, only AST modifications. Implementing a good verify is not so trivial. Bytecode in python 2/3 differs a little in some places. Furthermore we have too keep up with changes to it with newer versions of python.
The new unittests here (https://github.com/zopefoundation/RestrictedPython/tree/master/tests) are using a different approach to ensure that RestrictedPython does what it should. By really executing the code.

[tseaver]

I am fine with ditching the verify() call, as long as the new RP still allows that fibonacci to run as expected (with the tuple unpacking).

[tseaver]

I can't tell whether we have tests that tuple unpacking works: are the examples in RestrictedPython/tests/ still checked?

[icemac]

@tseaver The tests in src/RestrictedPython/tests are still run but only on the old implementation. We are about to rewrite them in the tests directory which are run on both implementations. There is also a test for tuple unpacking. (Even unpacking with star arguments is now supported, thanks to @stephan-hof!)