Compatibility with IA v1

[lbenet]

No description provided.

[Kolaru]

I had done some work on this as well, mostly through search and replace through the whole repository.

I have merge the two version and put it in the kolaru_IA1.0 branch. You may consider integrating to this one if you thing it is helpful (I don't have time to look in more detail for now).

The main difference is that:

• I use SVector internally for the models, and still parametrize the type on its length
• I used AbstractVector{<:Interval} to replace IntervalBox where possible

[lbenet]

Thanks Benoît for the efforts and the ideas; I like both of them. Currently, I'm now in the "validated" integration part, which I am moving to a submodule (it is sort of a separate thing), so if you want to push your suggestions, please do so.

[lbenet]

(I can also incorporate them, whenever I also find the time.)

[Kolaru]

I have added my changes. There are some tests failing that I don't have time to look at right now, but (normally) I fixed all type errors.

[lbenet]

Thanks a lot @Kolaru for including your changes; I'll take some time tomorrow (or the day after) to see what happens with the tests.

[schillic]

It seems that only a few docs entries are missing here, right? Would be great to get this merged and released!

[schillic]

I checked and the (very simple!) fix is to modify the line

TaylorModels.jl/docs/src/api.md

Line 7 in bcca0bb

Modules = [TaylorModels]

to

Modules = [TaylorModels, TaylorModels.ValidatedInteg]

[lbenet]

Once the tests are finishes, I'll merge this, leaving the remaining issues for a future fix...

Thanks a lot for the fix for the docs @schillic !

[schillic]

By the way, IA now already has v0.23 released. Maybe you can include v0.23 as well.

[lbenet]

Indeed, I forgot about that; I'm on this, but there are also issues of compatibility going back to TS, so I'm solving them, and then get back here...

[schillic]

I observed this persistent-task error recently in some of our packages as well. It may be unrelated (but I cannot say).

[lbenet]

I observed this persistent-task error recently in some of our packages as well. It may be unrelated (but I cannot say).

Are they also related to Base.UUID, or other packages?

[schillic]

Not sure, and I could not find where I saw this now. So just ignore my comment. In any case, this is only happening on nightly...

[lbenet]

After upgrading IA to v0.23 some problems have arosen, which are being solved in JuliaDiff/TaylorSeries.jl#386, but require some intervention in TaylorIntegration too...

[schillic]

Is there anything left to be done here?

(Maybe rename the issue title, by the way.)

[lbenet]

Is there anything left to be done here?

Something I'm working on, which I'd like to be in this PR, is related to the "NG" label in IntervalArithmetic; it would be nice to have that the guarantee flag is true. Also, I've been spending some time reanalizing the validated integration bounds, and other subtle points, which will not be in this PR (but that are based on it).

(Maybe rename the issue title, by the way.)

I've changed it, though perhaps in a week or so it will be ok :-)

[schillic]

May I ask that you either first make a release for v0.22 or also still allow the versions v0.22 and v0.23? Otherwise, all downstream packages must also jump directly from v0.20 to v1, which skips three main releases and may make the updates more difficult.

[lbenet]

Auch, I just tagged TaylorSeries which requires v1 in IA...

Once said this, there are minor changes between v0.22 and 0.23; before pushing this I checked locally that everything was ok. I can hold a bit to merge this, and perhaps use the previous tagged version of TaylorSeries, but IntervalRootFinding will also be tagged soon.

[schillic]

I do not see the problem: TaylorSeries has a previous release for v0.23, so Pkg can use that version.
Also, do you really need to require v1? The tests passed before, so I do not see a reason to not also allow v0.22 and v0.23, at least until the point where they become incompatible. It just creates more constraints on Pkg...

[lbenet]

Regarding TaylorSeries, in another commit I'll add also compatibility with v0.23, so everything works nice. Sorry for the mess, I'm too happy about IA v1.0

[lbenet]

Problem solved with the last release of TS (v0.20.7).

[schillic]

But this PR still requires v1 with the latest commit. I suggest to change it to IntervalArithmetic = "0.22, 0.23, 1".

[lbenet]

I think this is ready to be merged. Question @schillic : is it worth to tag a new minor version (as this PR proposes), or a patch version is enough?

[schillic]

When a reexported library had a breaking release, I usually make a breaking release as well. You know better if the breaking change would typically affect a user of TaylorModels.

[schillic]

I noticed that one actually cannot use TaylorModels with IntervalArithmetic v1 because IntervalRootFinding holds the version back. Any plan to support v1 there?

[lbenet]

I think v0.23 of IntervalArithmetic corresponds to v1.0; in that sense (and thanks to a suggestion you made!), it works. Once said that, it will not include the last changes in IA, i.e. v1.0.2. I opened a PR to address this (JuliaIntervals/IntervalRootFinding.jl#229).