Good news (so far) -- I tracked down the source of the bug. I had a long calculation (the details of which are irrelevant) which produced a pair of points on an elliptic curve, say P1 and P2, over a finite field of the form GF(2^n). In order to check the calculation I needed to check
Q1 == 2**(1+m)*P1 and Q2 == 2**(1+m)*P2 where Q1,Q2 were another pair of points, and m was some integer (actually calculated as len(x) where x was some list). I put in tracing in both the .py and .pyx version and saw that P1,P2, Q1,Q2 were the same in each version, but the comparisons weren't! I should say that in this case n was 251 and m 200. So far I can't get an isolated case like this where I choose P1 to be a random point, but according to the traces everything computed up to that point was the same. Victor On Aug 3, 8:38 am, VictorMiller <[email protected]> wrote: > Robert, I'll see what I can do. As you suspected, the files are not > disclosable :-(. > > Victor > > On Aug 3, 3:03 am, Robert Bradshaw <[email protected]> > wrote: > > > This is exactly the kind if thing we try to avoid. I'd like to see the > > files (if they're disclosable of course), or perhaps you could come up > > with a whittled-down example. > > > On Tue, Aug 2, 2011 at 7:22 PM, VictorMiller <[email protected]> wrote: > > > Robert, The .py and .pyx files are identical. I copied one to the > > > other, and just in case I checked with diff. It's very puzzling. > > > > Victor > > > > On Aug 2, 8:19 pm, Robert Bradshaw <[email protected]> > > > wrote: > > >> On Tue, Aug 2, 2011 at 11:29 AM, William Stein <[email protected]> wrote: > > >> > On Tue, Aug 2, 2011 at 11:09 AM, VictorMiller < [email protected]> wrote: > > >> >> I've written a bunch of functions (some organized in classes) to do > > >> >> some large computations in a particular finite field (always GF(2^n) > > >> >> for some odd n). This seems to work fine. I'd like the computation > > >> >> to be as fast as possible, so the first thing I did was to copy > > >> >> the .py file to a .pyx file. The good news is that the compiled > > >> >> cython is at least 30% faster (sometimes more) than the > > >> >> interpreted .py version. The bad news is that it gives different > > >> >> results! In trying to track down where things go awry, I made the > > >> >> following declaration in each > > > >> >> from sage.misc.decorators import sage_wrap > > >> >> from string import join > > >> >> def logged(func): > > >> >> @sage_wrap(func) > > >> >> def with_logging(*args, **kwds): > > >> >> print func.__name__ + '(' + join([str(_) for _ in args],',') > > >> >> + ')' > > >> >> return func(*args,**kwds) > > >> >> return with_logging > > > >> >> I then put > > >> >> @logged > > > >> >> in from of the defs of a bunch of functions. > > > >> >> This works as expected with .py version, but when I try to compile > > >> >> the .pyx version I get the message: > > > >> >> .... in update_wrapper > > >> >> setattr(wrapper, attr, getattr(wrapped, attr)) > > >> >> AttributeError: attribute '__doc__' of 'builtin_function_or_method' > > >> >> objects not writable > > > >> >> I think that I understand what's going on here, but are there any > > > >> > You often can't use decorators with Cython code, > > > >> yet... > > > >> > since it is compiled (not dynamic). > > > >> >> suggestions as to how to get to the bottom of the differences between > > >> >> the cython compiled version and the interpreted version? > > > >> > Put in print statements? > > > >> You could also write your own with_logging that doesn't try to access > > >> __name__ or __doc__ and decorate with that. > > > >> I'm very curious what goes awry. Did you make any changes other than > > >> changing the file from .py to .pyx and compiling it? > > > >> - Robert > > > > -- > > > To post to this group, send email to [email protected] > > > To unsubscribe from this group, send email to [email protected] > > > For more options, visit this group athttp:// groups.google.com/group/sage-support > > > URL:http://www.sagemath.org > > -- To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/sage-support URL: http://www.sagemath.org
