Author terry.reedy
Recipients terry.reedy
Date 2011-03-12.21:24:50
SpamBayes Score 3.74002e-11
Marked as misclassified No
Message-id <1299965091.63.0.267715976543.issue11477@psf.upfronthosting.co.za>
In-reply-to
Content
Example (which can serve as testcase with buggy output corrected).

class C(object):
    def __iter__(self):
        yield 'yes!'
    def __radd__(self, other):
        other.append('bug!')
        return other
    def __rmul__(self, other):
        other *= 2
        return other
    def __index__(self):
          return 3

class L(list):
    def __iadd__(self, other):
        list.__iadd__(self,other)
        return self
    def __mul__(self, other):
        return L(list.__imul__(self,other))

z1, z2, c = [], L(), C()
z1 += c
z2 += c
print(z1, z2, [1]*c, L([1])*c)
>>> 
['bug!'] ['yes!'] [1, 1] [1, 1, 1] # PyPy prints ['yes!'], [1, 1, 1]

Cause was diagnosed by Greg Price in
http://codespeak.net/pipermail/pypy-dev/2011q1/006958.html
as checking forward and reverse numeric slots before checking sequence slots for C-coded classes. Such a difference does not exist in Python itself.

In "About raising NotPortableWarning for CPython specific code"
pydev thread starting at
http://codespeak.net/pipermail/pypy-dev/2011q1/006958.html

Nick Coghlin identified the fix as "When a given operation has multiple C level slots, shouldn't we be checking all the LHS slots before checking any of the RHS slots?"

Guido replied "Yeah, indeed, on everything you said. The code dispatching based on internal slots is horribly ad-hoc and likely wrong in subtle ways."

I personally think fix should be backported to 2.7 and 3.2, but did not select them because that may be controversial.
History
Date User Action Args
2011-03-12 21:24:51terry.reedysetrecipients: + terry.reedy
2011-03-12 21:24:51terry.reedysetmessageid: <1299965091.63.0.267715976543.issue11477@psf.upfronthosting.co.za>
2011-03-12 21:24:50terry.reedylinkissue11477 messages
2011-03-12 21:24:50terry.reedycreate