The storage::snapshot() is used in two different modes -- one to save sstable as snapshot somewhere, and another one to create a copy of sstable. The latter use-case is "optimized" by snapshotting an sstable under new generation, but it's only true for local storage. Despite for S3 storage snapshot is not implemented, _cloning_ sstable stored on S3 is not necessarily going to be the same as doing a snapshot.
Another sign of snapshot and clone being different is that calling snapshot() for snapshot itself and for clone use two very different sets of arguments -- snapshotting specifies relative name and omits new generation, while cloning doesn't need "name" and instead provides generation. Recently (#26528) cloning got extra "leave_unsealed" tag, that makes no sense for snapshotting.
Having said that, this PR introduces sstables::storage::clone() method and modifies both, callers and implementations, according to the above features of each. As a result, code logic in both methods become much simpler and a bunch of bool classes and "_tag" helper structures goes away.
Improving internal APIs, no need to backport
Closesscylladb/scylladb#27871
* github.com:scylladb/scylladb:
sstables, storage: Drop unused bool classes and tags
sstables/storage: Drop create_links_common() overloads
sstable: Simplify storage::snapshot()
sstables: Introduce storage::clone()
567c28dd0d