There is a fundamental issue when adding nodes from versioning code right now.
Generally speaking, versioning code should not change after it has been written.
If old versioning code suddenly changes its behavior, it can invalidate the versioning
code that comes after it.
The issue is that while the versioning code itself doesn't change, it is indirectly calling
code that is changing and evolving all the time. This makes the versioning unreliable
and makes it hard to do certain changes (e.g. changing which sockets a node has).
It's easy to overlook issues this creates because many of these cases don't have
regression tests.
The solution is to keep the versioning code more independent from the part of
Blender that is constantly evolving. More specifically, this means that when adding
a node, this should not take the current node initialization and declaration code into
account. Instead, the versioning code should just manually create the node with
the sockets that it expects based on the version that the versioning code has been
written for.
This patch adds new `version_node_add_empty`, `version_node_add_socket` and
`version_node_add_link` methods which allow creating nodes without calling ever
evolving code.
Pull Request: https://projects.blender.org/blender/blender/pulls/121664
b210d56857