Add other notes

pycds/alembic/versions/434b4a868241_add_metadata_change_history/md_hx_100/other-notes.md

@@ -0,0 +1,55 @@
+# Other notes and questions
+
+## History table PK
+
+We could, possibly with some benefits, change the PK of the history table from the 
+current shared monotone serial `history_id` to a pair `(metadata_id, history_index)`, where
+`metadata_id` is the current metadata id (identifying a single item), and 
+`history_index` is a value that is serial (1, 2, 3, ...) for any given `metadata_id` 
+and gives a position in the history sequence for that one metadata item.
+
+This might have some value in enabling simple responses such as "How much history 
+does this item have?"
+
+## Table column ordering
+
+At present we reorder the columns of the primary table, inserting some history cols 
+(namely, `create_time`, `creator`) before them. This would make queries that
+depend on column order behave incorrectly.
+
+I think we can preserve the original column order and merely suffix new cols at the end. 
+
+## Avoiding duplication between primary and history tables
+
+We do not have to duplicate the latest record for each item; we can store the latest 
+state only in the primary and any past states in the history table. This makes full 
+history queries more trouble, since we will need to take the union of queries against 
+both tables. Furthermore, the present model has the primary and secondary table with 
+different columns. Possible responses:
+
+- Query against primary table supplements columns returned from it with additional 
+  computed columns to make it compatible with the history table.
+- Query against history table drops columns returned from it to make it compatible 
+  with the primary columns. This is **not workable** at minimum because the unique history 
+  FK would be missing.
+- Alter the primary table to have the same columns as the history table.
+
+### Alternative: Compute missing columns
+
+```postgresql
+SELECT *, FALSE AS deleted, ??? AS b_hx_id, ??? AS a_hx_id FROM b 
+UNION
+SELECT * FROM b_hx
+```
+Problems: 
+- `??? AS b_hx_id`: Need to compute the next hx id in `b` (or, if we adopt 
+  the above, the next hx index for that item -- probably harder).
+- `??? AS a_hx_id`: Need to compute current FK (or FK pair) for given `a_id` (ref). 
+  Not hard, already solved.
+
+### Alternative: Extend primary table columns
+
+Exposes things that might be better "hidden" in the history table.
+
+Makes the trigger functions somewhat more complicated and less well separated 
+functionally.