I believe your fix is correct.
The problem is that cse() is falsely assuming that it can replace a
Tuple with a Symbol. But this almost never works. You get similar
problems with other expressions that use Tuples in their .args. For
example:
In [612]: cse(meijerg((1, 2), (a, 4), (5,), [], x) + meijerg((1, 3),
(a, 4), (5,), [], x))
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
/Users/aaronmeurer/Documents/Python/sympy/sympy/<ipython-input-612-504251608d3e>
in <module>()
----> 1 cse(meijerg((1, 2), (a, 4), (5,), [], x) + meijerg((1, 3), (a,
4), (5,), [], x))
/Users/aaronmeurer/Documents/Python/sympy/sympy/sympy/simplify/cse_main.py
in cse(exprs, symbols, optimizations)
119
120 # Preprocess the expressions to give us better
optimization opportunities.
--> 121 reduced_exprs = [preprocess_for_cse(e, optimizations) for
e in exprs]
122 # Find all of the repeated subexpressions.
123 def insert(subtree):
/Users/aaronmeurer/Documents/Python/sympy/sympy/sympy/simplify/cse_main.py
in preprocess_for_cse(expr, optimizations)
41 for pre, post in optimizations:
42 if pre is not None:
---> 43 expr = pre(expr)
44 return expr
45
/Users/aaronmeurer/Documents/Python/sympy/sympy/sympy/core/exprtools.py
in factor_terms(expr, radical, clear, fraction)
566 list_args = [gcd_terms(a,
567 isprimitive=True,
--> 568 clear=clear,
569 fraction=fraction) for a in Add.make_args(p)]
570 p = Add._from_args(list_args) # gcd_terms will fix up ordering
/Users/aaronmeurer/Documents/Python/sympy/sympy/sympy/core/exprtools.py
in gcd_terms(terms, isprimitive, clear, fraction)
489 return a.func(*[gcd_terms(i, isprimitive,
clear) for i in a.args])
490 return type(a)([gcd_terms(i, isprimitive, clear,
fraction) for i in a])
--> 491 return gcd_terms(a, isprimitive, clear, fraction)
492 return terms.func(*[handle(i) for i in terms.args])
493
/Users/aaronmeurer/Documents/Python/sympy/sympy/sympy/functions/special/hyper.pyc
in __new__(cls, *args)
418
419 # TODO should we check convergence conditions?
--> 420 return Function.__new__(cls, tr(args[0]), tr(args[1]), args[2])
421
422 def fdiff(self, argindex=3):
/Users/aaronmeurer/Documents/Python/sympy/sympy/sympy/functions/special/hyper.pyc
in tr(p)
415 if len(p) != 2:
416 raise TypeError("wrong argument")
--> 417 return Tuple(_prep_tuple(p[0]), _prep_tuple(p[1]))
418
419 # TODO should we check convergence conditions?
/Users/aaronmeurer/Documents/Python/sympy/sympy/sympy/functions/special/hyper.pyc
in _prep_tuple(v)
26 """
27 from sympy.simplify.simplify import unpolarify
---> 28 return Tuple(*[unpolarify(x) for x in v])
29
30 class TupleParametersBase(Function):
TypeError: 'Integer' object is not iterable
Can you submit your patch as a pull request?
Aaron Meurer
On Sun, Apr 15, 2012 at 12:07 PM, Andreas Kloeckner
<[email protected]> wrote:
> Hi all,
>
> it looks like CSE and Subs don't interact very well. Here's what
> happens:
>
> - CSE finds a few Subs instances that share the entirety of their
> 'point' values.
>
> - CSE creates a symbol for the entire 'point' value--i.e. the whole
> Tuple.
>
> - Upon construction of the new Subs with the symbol for 'point', Subs
> wraps that symbol in a Tuple.
>
> - When the CSE'd value of the symbol is substituted back into the
> expression, we end up with Tuple(Tuple(expr)).
>
> There are multiple ways of fixing this, none strikes me as 'obviously
> right':
>
> - Special-case Subs in CSE, to only look at the elements of Subs.point,
> not the whole thing.
>
> - Special-case Tuple in CSE--in effect, don't collect entire Tuples in
> CSE. (*)
>
> - Denest in the Subs constructor, i.e. turn Tuple(Tuple(x)) into
> Tuple(x).
>
> - Add a flag to prevent the Tuple wrapping in Subs when called from CSE.
>
> (*) is what I've implemented as a stopgap, patch attached.
>
> I'd much appreciate any insight/fixes for this.
>
> Thanks,
> Andreas
>
--
You received this message because you are subscribed to the Google Groups
"sympy" group.
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/sympy?hl=en.
