Cap older versions breaking everything?

[galenlynch]

I've been pulling my hair out the past two hours because all of a sudden packages will not update, install, resolve, remove, anything...

I keep getting this error when using Pkg3:
restricted by julia compatibility requirements to versions: uninstalled — no versions left

When I inspected the Compat.toml of the offending packages, I saw that indeed, packages which had been working on 0.7/1.0 were now only accepting julia 0.6.

Then I noticed the most recent commit 0c75a4d, which made this change, intentionally?

So what am I supposed to do? I can no longer use the package manager. Am I not using pkg3 correctly, or is this a bug?

[galenlynch]

If I checkout the last commit before 0c75a4d, then I can use the package manager again.

[StefanKarpinski]

What are you trying to install? Without that information, we can't fix it.

[lloigor]

Mamba and Missings are broken for me (installing for 0.7).

[galenlynch]

Here are the packages that didn't work before I gave up: Decimals, ShowItLikeYouBuildIt, Destruct, Seaborn, TOML, PkgDev, but it felt like the tip of the iceberg. If I somehow got around one unsatisfiable requirement, it would just reveal another.

[KristofferC]

Do these work on 0.7?

[StefanKarpinski]

I'm able to dev Decimals and then test Decimals but it does print a lot of deprecation warnings.

[galenlynch]

I'm not sure if PkgDev did (I don't think it did), but Destruct and Seaborn certainly did. Looking at the released versions for the others, Decimals seems to have a release explicitly for 1.0 support, TOML seems like a maybe, ShowItLikeYouBuildIt seems doesn't have a version about compatibility.

I don't know, I've been using 0.7 for a month or more now without any problems, but now I can't. The only thing I tried to add this morning when I discovered the problem was JLD2, which I think is 1.0 compatible.

[StefanKarpinski]

• ShowItLikeYouBuildIt fails tests on 0.7 and 1.0 so that's a legitimate cap.
• Seaborn is broken and needs to use Test instead of Base.Test.
• Destruct also has deprecations in its tests.
• etc.

In short, I think the answer to this is mostly that people need to fix their shit. I'm still perplexed that people are so unwilling to just fix a few deprecations, but now is the time to do so. If you want to work around this while people get their acts together and fix things and make releases of packages that aren't broken, you can go into ~/.julia/registries/General and do

git checkout -b dontcapmebro e5748a7e36745cf42952c5451586a79911a524b4

[galenlynch]

It might be true that people need to fix their packages, but as a user yesterday Julia worked, and today it doesn't.

I thought the point of 0.7 was that it was 1.0 with deprecation warnings instead of failure. Maybe you could change the enforced cap to include 0.7?

[KristofferC]

Maybe you could change the enforced cap to include 0.7?

I've had the same thought.

[StefanKarpinski]

Sure, we could potentially do that. I don't really care about 0.7, so 🤷‍♂️

[galenlynch]

Dunno, I'll gladly switch to 1.0 once the packages I need to graduate stop throwing deprecation warnings. Until then, 0.7 it is.

Not that I get a vote, but if I did I would vote for relaxing the cap to allow minimally maintained, but still useful, packages to work and be installable on 0.7.

[timholy]

I can verify that there was a problem but that 1a5454c seems to have resolved the test I was about to report.

[galenlynch]

Thanks for relaxing the cap, it makes life much easier!

[StefanKarpinski]

Sure, happy to make life easier while still having correct caps for 1.0 at least.