On Fri, Mar 6, 2015 at 2:09 PM, Joachim Durchholz <[email protected]> wrote: > Hi all, > > I've been working on removing C, and now that all the simple cases are done, > I'm getting to the harder ones - those where SymPy code is called during the > import (which will fail if the SymPy code being called has not yet fully > imported). > > There's the solvable cases - stuff like ask.py's known-properties database, > where the initialization can be moved from module-level code to a function, > so the initialization will happen on demand, after the "import sympy" > statement has completed. > > And now I'm at an unsolvable case: anything that uses S will run class > initialization code. And replacing that code with something delayed would be > impractical. > > Here's the bottom end of such a stack trace: > > File "sets/fancysets.py", line 173, in <module> > class Reals(with_metaclass(Singleton, Interval)): > File "core/compatibility.py", line 181, in __new__ > return meta(name, bases, d) > File "core/singleton.py", line 70, in __init__ > the_instance = ctor(cls) > File "sets/fancysets.py", line 176, in __new__ > return Interval.__new__(cls, -S.Infinity, S.Infinity) > File "sets/sets.py", line 744, in __new__ > if (end < start) == True: > File "core/numbers.py", line 1005, in __lt__ > other = other.evalf()
I think ideally we should avoid calls like this being run at import time. To me, this indicates a design issue in the sets. Aaron Meurer > File "core/evalf.py", line 1343, in evalf > _create_evalf_table() > File "core/evalf.py", line 1184, in _create_evalf_table > from sympy.functions.combinatorial.numbers import bernoulli > ImportError: cannot import name bernoulli > > What happens is that the "Reals" class gets initialized, which is a > Singleton, so its Singleton metaclass runs the constructor to get "the > instance": > File "core/singleton.py", line 70, in __init__ > the_instance = ctor(cls) > In the case of Reals, the constructor quite reasonably tries to define th > set of Reals as the Interval between -Infinity and Infinity. Unfortunately, > this runs SymPy code during class definition time, which is during import > time, at which time we can't rely on SymPy being fully available yet. In > this particular instance, the code runs into _create_evalf_table, but minor > restructuring could shift the actual point of failure almost anywhere - all > that's needed is an import that runs some piece of SymPy code that isn't > imported yet. > > > Question is: WHAT DO WE DO? > =========================== > > 1) What *will* work is to replace the attribute with a function. > Attributes need to be initialized before use, functions don't need to work > until they are actually needed. > Except then S.Foo would have to be rewritten as S.Foo() for all values of > Foo. Anybody who ever used SymPy and picked up the S.Foo idiom for his code > will curse us to hell if we do that. > > 2) We could defer creation of the_instance. S would first simply collect > names for its namespace, and trigger instance creation on request (which > could be done from sympy.__init__.py). > Any uses of S at import time would fail and need to be fixed (except those > in the Singleton class constructors). I do not know how many files would be > affected by that -> not nice, but maybe an option. > > 3) We could change the code that accesses the_instance so that actual > construction is delayed until actual use. > the_instance is used only inside the __init__ function of Singleton: > > class Singleton(ManagedProperties): > ... > def __init__(cls, name, bases, dict_): > ... > # The creation call that needs to be delayed: > the_instance = ctor(cls) > def __new__(cls): > # Easy: call ctor(cls) on the first __new__. > return the_instance > ... > # Harder: route S's getattr to a function that > # runs ctor(cl) to construct the_instance. > setattr(S, name, the_instance) > > I don't have a full grasp of what ManagedProperties does yet, so this may or > may not be doable. > > > Questions? Comments? More ideas how to deal with this? > > -- > You received this message because you are subscribed to the Google Groups > "sympy" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at http://groups.google.com/group/sympy. > To view this discussion on the web visit > https://groups.google.com/d/msgid/sympy/54FA0980.9070304%40durchholz.org. > For more options, visit https://groups.google.com/d/optout. -- You received this message because you are subscribed to the Google Groups "sympy" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/sympy. To view this discussion on the web visit https://groups.google.com/d/msgid/sympy/CAKgW%3D6LUNJBXnxrObzhcTz%3DaLBi9%3Dc0NNqFw05W%2ByKJoPjznEQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
