On Wednesday, January 28, 2015 at 12:03:00 AM UTC-8, vdelecroix wrote:
> This is really annoying from the user perspective
Not nearly as annoying as the unavoidable memory leaks we had before,
though. Note that most internal structure types do not suffer from
slowdowns, so you could study them to see what the difference is for
keeping those homsets alive.
(and it might also
> be related to the infinitely longer TestSuite(M) that I mentioned
> above).
Quite possibly. Profiling would be quite useful.
Do you have an idea for a mechanism that would make this
> automatic ?
> I mean while doing
>
> sage: P = ProductRing((GF(5), GF(5)))
>
> I would like that all important coercions to be kept alive until P
> dies. It might be something explicit in the constructor of
> ProductRing. Would the following of good practice within Sage:
>
> class ProductRing(Parent):
> def __init__(self, ...):
> ...
> self._cache = [self.coerce_map_from(GF(5))]
>
No. The coercion system has a cache of its own. It does keep strong
references to the coercion maps. They don't have homsets, though. The
problem might be that you keep extracting coercion maps from the coercion
system:
if all(R.has_coerce_map_from(S) for R in self._rings):
return CallableConvertMap(S, self,
lambda x: self.element_class(tuple(R.coerce_map_from(S)(x)
for R in self._rings), parent=self),
parent_as_first_arg=False)
If you rewrite this as:
if all(R.has_coerce_map_from(S) for R in self._rings):
coerce_maps = tuple( R.coerce_map_from(S) for R in self._rings)
return CallableConvertMap(S, self,
lambda x: self.element_class( tuple(m(x) for m in
coerce_maps), parent=self),
parent_as_first_arg=False)
is a lot faster. Now the lifetime of the homsets is tied (via storing them
in a closure) to the lifetime of the coercion map that needs them, and the
coercion system itself can cache that map for as long as needed. There are
other ways of writing this that are probably even faster. I would imagine
that
lambda x: self.element_class( tuple( R(x) for R in self._rings),parent=self)
would be even faster (and possibly more memory efficient, because now
everything is left to the coercion system instead of trying to extract
information from the coercion system in the form of maps).
--
You received this message because you are subscribed to the Google Groups
"sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to sage-devel+unsubscr...@googlegroups.com.
To post to this group, send email to sage-devel@googlegroups.com.
Visit this group at http://groups.google.com/group/sage-devel.
For more options, visit https://groups.google.com/d/optout.
