Reconsidering the table inheritance methods used

As according to Martin Fowler, versioncontrol makes heavy use of the highly normalized Class Table Inheritance pattern, where each "class" (in our case, array) gets its own table and only contains fields relevant only to that class. For example, versioncontrol_operations is the supertype class, and versioncontrol_commits, versioncontrol_branch_operations and versioncontrol_tag_operations are the subclasses. In order to access all of the data for a JOIN needs to be performed; in practice two JOINs are necessary due to the fact that each repository can add its own data.

I would like to bring up for consideration the Single Table Inheritance pattern, which would abolish those subclass tables and put everything in versioncontrol_operations. Here's why:

• The subclass tables contain very few columns, so space inefficiency would largely be a non-issue; it might even be more efficient because we remove the need for foreign keys
• It simplifies queries by eliminating the need for a second join beyond the join we would need to link this table with the repository-specific one
• There are duplicate columns in some of the tables, for example description is present in commits and tags

There would only be a slight denormalization; at most two columns would not be used by a datatype. Backwards compatibility can be maintained using views to emulate the older tables. Thoughts?