Tagging a minor release

[SylvainCorlay]

We are about to make a first public release of a (open-source) package that makes use of pybind and depends on a few things that are currently in master. However we would need a release including these features?

Would you be ok with tagging a v1.8.2 or v1.9?

^ @wjakob

[jagerman]

I think we're definitely in almost-at-2.0 territory (#442), so maybe v2.0.0-beta1 or some such?

[wjakob]

Yes, it will need to go in v2.0. We're currently still blocked waiting for some NumPy related changes making it into master before the release.

[SylvainCorlay]

Arg, this is blocking the release of xtensor-python.

Is the current state of master broken so that you can't make a minor release first? A beta release cannot go into conda-forge.

[SylvainCorlay]

@jagerman btw, have you checked out http://quantstack.net/xtensor ?

[wjakob]

There are tons of breaking changes, so making anything non-v2.x would break semver. It might be possible to back-port selected changed, but this would be a lot of work.

[SylvainCorlay]

Have we been following semver since 1.0?

[wjakob]

cc @aldanor

[wjakob]

Have we been following semver since 1.0?

Not since 1.0, but we just promised that we would follow it with the last release.

[wjakob]

BTW: If it's such a big deal to depend on the latest version of pybind11, can't you just use the git master as a pip/conda dependency? At least with Pip I know that you can do that.

[SylvainCorlay]

The public release of xtensor is a good occasion to grow the ecosystem around pybind.
I feel that it is likely that issues in pybind will come up as more people use xtensor, and we might want to iterate on this before a 2.0 rather than after.

It seems that conda does not support pip dependencies in recipes (only in environments)

For the actual 2.0, you may also want to make the eigen bindings an extension within the pybind org rather than a core pybind11 feature.

[jagerman]

@SylvainCorlay Just had a look: looks like a nifty project.

[SylvainCorlay]

@SylvainCorlay Just had a look: looks like a nifty project.

We would love to have some feedback from you guys! As well as @aldanor .

[wjakob]

@SylvainCorlay : Hm, those are some pretty strong suggestions. I don't think I'm ready to kick out Eigen from the main repo since it's such a widely used package in scientific computing. (Also for selfish reasons, all of my projects use it ;))

I feel that it is likely that issues in pybind will come up as more people use xtensor, and we might want to iterate on this before a 2.0 rather than after.

I did not understand what you meant with this. Are you saying we should backport non-breaking changes to v1.x and do a release? This is in principle possible, but it will require a lot of cycles that I unfortunately don't have in the short term.

[SylvainCorlay]

I feel that it is likely that issues in pybind will come up as more people use xtensor, and we might want to iterate on this before a 2.0 rather than after.

I did not understand what you meant with this. Are you saying we should backport non-breaking changes to v1.x and do a release? This is in principle possible, but it will require a lot of cycles that I unfortunately don't have in the short term.

Ah that is not at all what I meant :) I meant that when xtensor is released, we are likely to hit the new APIs a lot more and find issues.

If we wait for pybind 2.0, it is likely that we will have to make multiple patch releases of pybind to fix those... If instead we go through this phase rapid iteration before 2.0 and release 2.0 when it is stabilized, the major release number will convey stability.

[jagerman]

Re: Eigen, me too.

That said, I could see some potential for some future Eigen support that goes far further than the current one, and if it came to that, putting it in a separate project would make sense; but for now, eigen.h is literally just 4 type casters around Eigen types (well, okay, template types). It's good enough to shuttle data between Eigen and numpy, but it really isn't a proper Eigen interface (nor does it claim to be).