#15367: Empty lists while creating parents
-------------------------------------+-------------------------------------
Reporter: roed | Owner:
Type: defect | Status: new
Priority: major | Milestone: sage-5.13
Component: memleak | Resolution:
Keywords: | Merged in:
Authors: Nils Bruin | Reviewers: Simon King
Report Upstream: N/A | Work issues: Improve timings
Branch: | Commit:
u/SimonKing/ticket/15367 | a2852e9610a76e6df392544b01b8e3c319b1cdd4
Dependencies: | Stopgaps:
-------------------------------------+-------------------------------------
Comment (by nbruin):
Ouch, that is indeed not such good news and with such a degradation in
performance I would definitely not recommend merging it. Things I can
think of:
- We might have to improve the initial seed of the hash function. If
we're getting many collisions, For `dict`, `oldMonoDict` and `newMonoDict`
we're all using different hash functions. In the test above you never
timed pure data item retrieval (both existing and non-existing entries).
If we can have that data too, we might be able to see if the problem is
there
- We are probably hitting resizing more often (since dict starts with
quadrupling, it resizes less), so if that's a problem we could be more
aggressive there.
- I am allocating memory on delete as well, for well-documented reasons.
That's more expensive than straight dereffing. Maybe much more expensive?
- perhaps the GC code around the new dicts is less than optimal? I
checked the C-code that Cython generated for the `_traverse` and `_clear`
methods and was pretty happy with them, but perhaps there's overhead there
as well.
Finally, note that `MonoDict` and `TripleDict` are optimized for read
speed, not particularly the other operations. I think that's the frequent
operation on them in the coercion framework; discovery much less so
(otherwise parents must be disappearing/reappearing all the time and the
creation overhead of the parent is likely much higher).
--
Ticket URL: <http://trac.sagemath.org/ticket/15367#comment:24>
Sage <http://www.sagemath.org>
Sage: Creating a Viable Open Source Alternative to Magma, Maple, Mathematica,
and MATLAB
--
You received this message because you are subscribed to the Google Groups
"sage-trac" 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/sage-trac.
For more options, visit https://groups.google.com/groups/opt_out.
