On 2023/12/20 19:44, Daniel Dickman wrote:
> > 
> > If I read things correctly, it's anyio that has the tight version spec,
> > so there should be a mention in there. (However I don't see that reflected
> > in jupyter_server's pyproject file, which just says anyio>=3.1.0...)
> 
> I assume you're looking at a more recent version of jupyter_server. For 
> example I see the latest on pypi is 2.12.1.

Ah, that explains it, I was looking at the most recent pyproject.toml in
git.

> Unfortunately we're so far behind with our jupyter ports that the version 
> I'm looking to bring in to start is 1.8.0 from May 2021. That version has:
> 
>   $ grep -n anyio setup.cfg
>   43: anyio>=3.1.0,<4
> 
> I wasn't able to work out how to go straight to all the latest jupyter 
> versions and it looks to me like updates may need to be done in small 
> steps.
> 
> One of the biggest pain points is going to the next version of nbconvert 
> will require us to add nbclient as a new dep which seems to result in 
> various circulat dependencies. And that seems to be a headache because I 
> think Python is able to handle circular dependencies, while our ports tree 
> cannot handle ciruclar RDEPs.

We can however place a dependency in what is technically the wrong port
though. Say X requires Y and vice-versa but in practice they're only
really used by Z - we can put the dep on both X and Y within Z and that
will be good enough for real world use.

> I haven't worked out if it'll even be possible to upgrade jupyter fully in 
> our tree, but essentially the more we can update with updating nbconvert, 
> the better as a start.
> 
> As a first step I was able to get jupyter notebook updated to 6.5.6 (the 
> last version before the 7.x series which makes things a bit tricky). So 
> let me start by sending that to ports@ shortly...
> 
> I also did the work to get jupyterlab 3.2.9. I'll send that out as well.
> 

Reply via email to