Dynamic software updating research efforts have mostly been focused on updating application code and in-memory state. As more and more applications use embedded databases for storage, dynamic updating solutions will have to support changes to embedded database schemas. The first step towards supporting dynamic updates to embedded database schemas is understanding how these schemas change - so far, schema evolution studies have focused on large, enterprise-class databases. In this paper we propose an approach for automatically extracting embedded schemas from regular applications, e.g., written in C and C++, and automatically computing how schemas change as applications evolve. To showcase our approach, we perform a long-term schema evolution study on four popular open source programs that use embedded databases: Firefox, Monotone, BiblioteQ and Vienna. Our study spans 18 cumulative years of schema evolution and reveals that change patterns and frequency in embedded databases differ from schema changes in enterprise-class databases that formed the object of prior studies. Our platform can be used for performing long-term, large-scale embedded schema evolution studies that are potentially beneficial to dynamic updating and schema evolution researchers.