#14711: Memleak when creating QuadraticField
-------------------------------------------------+-------------------------
Reporter: jpflori | Owner:
Type: defect | davidloeffler
Priority: critical | Status: new
Component: number fields | Milestone: sage-5.13
Keywords: memleak, number field, | Resolution:
QuadraticField | Merged in:
Authors: | Reviewers:
Report Upstream: N/A | Work issues:
Branch: u/SimonKing/ticket/14711 | Commit:
Dependencies: | Stopgaps:
-------------------------------------------------+-------------------------
Comment (by SimonKing):
To summarise the requirements found in our discussion for a stable and
memory
friendly coercion system:
- Maps registered by `P.register_coercion(mor)` are the backbone of the
discover_coercion algorithm. Hence, they need to be kept healthy as long
as `P` lives.
- If P has a coerce embedding then this embedding needs to be kept healthy
as long as P lives.
- If the coercion system finds a coercion from P to Q different from the
two preceding cases (i.e., by transitive closure), then P must not prevent
Q and the coercion map from garbage collection, and Q must not prevent P
and the coercion map from garbage collection. But the coercion map needs
to stay healthy as long as both P and Q are alive. In particular, take
care of composed maps and actions.
To summarise how I think we can meet these requirements:
- `P.register_coercion(mor)` keeps a strong reference to `mor.domain()` in
a new attribute `P._registered_domains` (which is a list). All other maps
in the coercion system will be weakened.
- Homsets keep strong references to domain and codomain.
- "Weakening a map" means to use a weak reference to the domain and a
removed reference to the homset. But we will keep a strong reference to
the codomain. Consequence: (1) In a composed map or action, the "middle
parents" will always stay alive as long as the composed map lives. (2)
Embeddings do what we required above: The embedding keeps the codomain
alive.
- We store the map phi found by `P.discover_coercion(Q)` in
`P._coerce_from_hash`, which is a monodict. In particular, if Q will be
garbage collected, then phi is immediately removed from the cache, and
thus the strong reference of phi to the codomain P will not prevent P from
garbage collection. And if P will be garbage collected, then the whole
cache including phi is removed. However, if both P and Q live, then phi
will stay in `P._coerce_from_hash[Q]`, and it will stay healthy, because
its domain Q is still alive (and hence the weak reference is still ok).
I think this model makes sense.
And concerning a safe user interface: Perhaps we can do without. Namely,
we could make it so that the string representation of a weakened map will
consist of a warning, suggesting to use a copy. In this way, it would be
sufficiently likely that the user wouldn't have too many bad surprises.
--
Ticket URL: <http://trac.sagemath.org/ticket/14711#comment:82>
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.
