Speed up coercions

[mezzarobba]

• Don't call _repr_() on the parents every time a coercion or conversion fails. (One could consider making Parent.__repr__() keep the name in cache instead, but we probably don't want that for things like finite fields, and in any case __repr__() is handled by SageObject at the moment.)
• Add a fast path for trivial coercions in Parent.coerce(foo).

Before:

sage: R.<x,y,z> = ZZ[]
sage: a = GF(3)(2)
sage: %timeit x(a,a,a)
The slowest run took 37.44 times longer than the fastest. This could mean that an intermediate result is being cached.
100000 loops, best of 3: 13.1 µs per loop

After:

sage: R.<x,y,z> = ZZ[]
sage: a = GF(3)(2)
sage: %timeit x(a,a,a)
The slowest run took 162.14 times longer than the fastest. This could mean that an intermediate result is being cached.
100000 loops, best of 3: 3.98 µs per loop

Component: performance

Author: Marc Mezzarobba

Branch/Commit: 3b49d0a

Reviewer: Travis Scrimshaw

Issue created by migration from https://trac.sagemath.org/ticket/27379

[tscrim]

comment:2

LGTM.

I do not think we want to cache the result of Parent.__repr__ because that is designed so you can dynamically change the repr of the object via foo.rename('bar').

[tscrim]

Reviewer: Travis Scrimshaw

[mezzarobba]

comment:3

Replying to @tscrim:

LGTM.

Thank you!

I do not think we want to cache the result of Parent.__repr__ because that is designed so you can dynamically change the repr of the object via foo.rename('bar').

Yes, that's an additional reason, though one could imagine making rename() update the cache.

[vbraun]

Changed branch from u/mmezzarobba/optimize_coerce to 3b49d0a