As they are currently implemented, aliases have a particular edge case which should be handled, but are not.

When there is a column, for example id, which exists in multiple tables but do not refer to the same id and a user wants one instance of id to be in an alias (and ignore others), this is not currently supported.

Background

In the current iteration of this utility, the user must know about the order with which to deidentify columns. Let's take a simple example:

If dataframe has two columns, id and timestamp, the two associated maps would be:

id_map: id -> new_id
time_map:id -> new_timestamp

OR

id_map: id -> new_id
time_map: new_id -> new_timestamp

The difference between these two makes a large difference to their usage.

In case 1, the user must shift the time columns BEFORE shifting the id (or else the time map will not see the associated IDs).

In case 2, the user must shift the id first BEFORE shifting the times.

This is an issue that needs to be resolved in order for ease of use for the end-user.

Early thoughts:

the two major paths this could take is to create an entity that has both the logic for the time AND the id in the same object, or to somehow make it clearer which order must be used.

Users need to be able to specify all of the files they want to deidentify with which maps and then have a script that loops over all of the files and applies the correct mappings

When the configuration is autogenerated, it can have artifacts due to .csv escaping behavior, i.e.

file1:
    id:
        '"column1"'

This should be handled in the generation stage or afterwards. Probably in the generation stage.

Need to implement a way for the offset map to update when the underlying time offset method is moving to the start year. This may get confusing, as if this means if the currently DataFrame being processed is not of the same nature of timestamp as the original one used to create the offset_map, unintended behavior may occur.