[SciPy-Dev] SciPy and pybind11 2.7

[Drew Parsons]

On 2021-08-17 20:16, Ralf Gommers wrote:
> On Tue, Aug 17, 2021 at 7:02 PM Drew Parsons <dparsons at debian.org>
> wrote:
> 
>> Hi SciPy developers,
>> 
>> the release version of SciPy gives upper version caps on
>> dependencies,
>> e.g. pybind11>=2.4.3,<2.7.0
>> 
>> The latest pybind11 release is now 2.7.1.  Is there a specific
>> reason
>> why SciPy adds the version cap to releases, i.e. a specific known
>> point
>> of failure?  Or is the version cap simply used conservatively to
>> help
>> maintain release stability for users?
> 
> There is no specific problem, it's just a matter of controlling the
> upgrade timeline so there is no risk of a new pybind11 release
> breaking an already released version of SciPy. Because it's a
> build-only dependency, that is usually the right strategy (runtime
> dependencies are a bit different, because the upper bound may bother
> users).
> 
> If you want to bump the version in master, please let us know or open
> a PR.
> 


Thanks Ralf. I had a feeling that in principle scipy could build fine 
with pybind11 2.7, since the build wasn't reporting any errors. I just 
wanted assurance that there weren't any known but unseen problems.

My build system (Debian) can happily ignore the recommended upper limit, 
so I've now uploaded scipy 1.7.1 to debian unstable, see 
https://buildd.debian.org/status/package.php?p=scipy

Hence, no urgency on my account to bump the version in master, do it 
once you're comfortable that scipy is fine with pybind11 2.7. We can 
consider Debian unstable as a test environment for it :)

Cheers,
Drew