rip out orphan logic?

[StefanKarpinski]

Came up here: #4152 and thought I'd open a separate issue to discuss getting rid of all the orphan logic for Pkg GC:

FWIW, I think the whole orphan logic was a mistake and we could delete it all. The idea was that we didn't want to end up deleting things which then immediately need to be reinstalled, but it complicates things, makes GC not behave the way people expect, and I now highly doubt that the motivating problem actually would happen much.

[KristofferC]

I agree.

[IanButterworth]

For me part of why it's been a bit hard to understand is the list of manifests that it considers to be active isn't easily understandable nor queryable or editable.

I think that needs to be fixed with or without the orphaning mechanism, otherwise without the ophaning mechanism what would GC throw away? Everything not used by the current env?

[KristofferC]

Everything not used by the current env?

There will still be a list of active manifest. The orphan thing that is being discussed is that even if a manifest is removed so that some packages are not referred to by any tracked environments, the packages get tracked at a different location in case the packages will get reinstalled again and not immidiately GCed.

[IanButterworth]

Yeah. I'm saying removing the orphaning thing and exposing a way to query/edit the list of tracked environments sounds like a nice improvement together.
It might avoid/soften confusion if without the orphaning people start seeing stuff being unexpectedly removed.

[KristofferC]

without the orphaning people start seeing stuff being unexpectedly removed.

But why?

[IanButterworth]

A project gets moved/renamed, or a manifest or artifacts file specifically gets deleted.

The Pkg.gc() docs just state

Garbage-collect package and artifact installations by sweeping over all known Manifest.toml and Artifacts.toml files, noting those that have been deleted.

I'm not saying improving how to know what all known Manifest.toml and Artifacts.toml file includes is a blocker for removing the orphaning thing, just that it'd be a good addition in the same release.

For instance the news might read:

Pkg now tidies up unused stuff more eagerly, and exposes the following ways to query which environments it considers to be active ...

Currently the user has to know that .julia/logs/manifest_usage.toml etc. is a thing. Docs don't mention it (apart from a passing mention in the glossary), and Pkg.gc() doesn't name the files where its tracked.

(@v1.11) pkg> gc
      Active manifest files: 125 found
      Active artifact files: 253 found
      Active scratchspaces: 11 found
     Deleted no artifacts, repos, packages or scratchspaces

[IanButterworth]

Related: #2688

[DilumAluthge]

I agree that better visibility would be nice.

I may be misremembering, but does https://github.com/giordano/PkgCleanup.jl expose some of that information? cc: @giordano

[giordano]

The package only accesses the manifest usage https://github.com/giordano/PkgCleanup.jl/blob/a2518063016971e484bf790e5a2babef6a68dedb/src/PkgCleanup.jl#L101, not the orphaned.toml file.

[KristofferC]

A project gets moved/renamed, or a manifest or artifacts file specifically gets deleted.

Yeah, but unless you specify Seconds(1) or something for the GC time limit, it won't get deleted until the default time which is 7 days.