Follow-up from "Defer workspace source commits"

The following discussion from !1628 (merged) should be addressed:

• @juergbi started a discussion: (+2 comments)

  We no longer use sourcecache itself anymore, as far as I can tell. I'd suggest replacing the above with cas = self._get_context().get_cascache().

• workspace plugin should not access private API
• rework central source handling to resolve the issues identified by workspaces:

  we need to set the _key property of that source before we do any caching so we can accurately ref distinct trees and we need to allow logic for the workspace possibly not being cached in the usual workflow since it essentially is not if it has diverged from the checkout state

cc @BenjaminSchubert @juergbi

we should re-write part of this plugin to be closer to what the 'local' kind does, and we therefore wouldn't even need the cas anymore

Do you mean we should calculate the digest without actually importing the files into CAS at that point? This would work against the planned optimized import via casd: BuildGrid/buildbox/buildbox-casd#33 (moved)

Ah, I forgot about this one, yes if we can leverage buildbox-casd, then it's better, however we'll need to make some things public in order to keep it nice, or add a new API to the source plugin. (And then move the local plugin too). Did you have something in mind on how it will look like afterwards?

I agree, the goal should be to make this work without having the workspace plugin access internal API, and then we should aim for the same approach for the local plugin. However, I haven't looked into this in detail yet. And workspace handling is already closer to regular sources, so we're moving in the right direction.

assigned to @traveltissues

mentioned in issue #985 (closed)

changed the description

One more note here, I think most of the changes should also be applied to the 'local' source plugin, which should probably behave exactly as the workspace one does

We could implement a public api and/or move some of the logic from mrs like !1628 (merged) to source but it occurs to me that it would probably make more sense to address the way sources are handled which are the causes of the failures we're seeing eg. sources are possibly not always cached, etc

mentioned in merge request !1651 (merged)

added Enhancement Refactoring labels

closed via merge request !1651 (merged)

mentioned in commit e0e20993