Upgrade Arb to 2.14.0

[fredrik.johansson]

​https://github.com/fredrik-johansson/arb/archive/2.14.0.tar.gz

Because of more accurate error bounds (mainly for trigonometric functions), a large number of doctests have to be updated.

[fredrik.johansson]

New commits:

​d22edee	update arb to 2.14.0
​6f0646b	update doctests affected by accuracy changes in arb 2.14
​d4295f0	Merge branch 'arb214' into t/25966/upgrade_arb_to_2_14_0

[tscrim]

Buildbots ahoy! *cough*authorname*cough*

[vbraun]

**********************************************************************
File "src/sage/rings/complex_arb.pyx", line 1029, in sage.rings.complex_arb.ComplexBallField.?
Failed example:
    C.integral(lambda x, _: (x + x.exp()).sin(), 0, 8) # long time
Expected:
    [0.34740017...55347713 +/- 6.72e-598]
Got:
    [0.347400172657247807879512159119893124657456254866180183885492713616748213988785320529685104346604105756813796172006018707302714228197618073704004353678452866627436279471970216410908716043577412909956068587776704710948691279593004567821091508925369957240639547298886452332685264381903039098870525160076005978168808062746564134987731050142119306346307812114504420398584159417131096900798039662018221641127266271301569159908541386062975763888813212164702872661085356464323603726722886936732008463756755254875678750234857309302489895429562039632272592695502596880447563060128384223515181416786355347713 +/- 5.71e-598]
**********************************************************************
1 item had failures:
   1 of  56 in sage.rings.complex_arb.ComplexBallField.?
    [622 tests, 1 failure, 3.69 s]
----------------------------------------------------------------------
sage -t --long src/sage/rings/complex_arb.pyx  # 1 doctest failed
----------------------------------------------------------------------

[tscrim]

I must have done the same thing and not run --long. Mea culpa. Fixed.

---

New commits:

​61e2d8e	Merge branch 'u/fredrik.johansson/upgrade_arb_to_2_14_0' of git://trac.sagemath.org/sage into u/fredrik.johansson/upgrade_arb_to_2_14_0
​8bef4fd	Fix one doctest.

[fredrik.johansson]

Am I allowed to give the positive review to the reviewer's patch? Are we waiting for other reviewing?

[tscrim]

Replying to fredrik.johansson:

Am I allowed to give the positive review to the reviewer's patch?

Yes.

Are we waiting for other reviewing?

Not that I am aware of.

[fredrik.johansson]

Unfortunately Sage fails to build for me right now (for unrelated reasons) so I can't verify that the tests pass.

[gh-timokau]

Tests pass.

[gh-timokau]

In the future it would be preferable to make use of # abs tol 1e-42 feature to reduce the doctest precision. It would make the doctests less brittle.

[mmezzarobba]

Replying to gh-timokau:

In the future it would be preferable to make use of # abs tol 1e-42 feature to reduce the doctest precision. It would make the doctests less brittle.

See the discussion at #18560. IMO it would be time to switch to using tolerances indeed. Jeroen, did you view of the issue change since that ticket?

[fredrik.johansson]

Differences when upgrading Arb are not a big issue IMO. In fact, it's helpful (for me) to look at the Sage doctests to see what improved or got worse between two versions. However, divergence between 32-bit and 64-bit results is plausible in the future and tolerances may be the best solution for that.

[gh-timokau]

Replying to fredrik.johansson:

Differences when upgrading Arb are not a big issue IMO. In fact, it's helpful (for me) to look at the Sage doctests to see what improved or got worse between two versions. However, divergence between 32-bit and 64-bit results is plausible in the future and tolerances may be the best solution for that.

They are an issue when using sage outside of sage-the-distribution. They make it necessary that either the arb version used is the exact arb version sage-the-distribution uses or that patches are applied to sage. With tolerances we could potentially update sage and arb independently. And this is not only an issue for distributions. It could in the long run also allow sage-the-distribution to use the system's arb version (#24919).

In my opinion ideal sage tests should only fail if something is actually broken in sage, upstream projects or packaging. "broken" here may be an actual error or a api change.