deletes and table drops don't always cascade in correct order about datajoint-matlab HOT 4 CLOSED

Do you have an example? It's hard to reason about this without seeing a schema where the cascading fails.

from datajoint-matlab.

yes, deleting from reso.Align in the schemas from atlab/commons repository. But here is the minimal example of dependencies that will reproduce this:

A
B->A
C->B
D->A
D->C

If you delete from A, the algorithm will attempt to delete in sequence CDBA, which will fail because D should be deleted before C, on which it depends.

To solve this, we need to keep track of the tree depth. The tree depth is already computed in dj.Schema.tableLevels. So it's easy to fix when the cascades do not cross schema boundaries. But additional logic needs to be added or a new tree level needs to be computed for cases when the cascade crosses schema boundaries.

from datajoint-matlab.

This has been happening to me a few times recently and I've been wondering whether it's a bug in DJ or some inconsistency in my database. Good to know it's most likely not my db.

The crossing of schema boundaries should work as well in my opinion. Otherwise we need to be more explicit about the consequences and disadvantages of using multiple schemas.

from datajoint-matlab.

Additional fix in dc707af. Would be great to get some testing feedback.

from datajoint-matlab.