Re: [xwiki-devs] [Proposal] Use liquibase to manage the database schema

While trying to reduce the likelihood of duplicate ids for documents, and extending my patch to provide a proper solution for objects, I fall on a really unexpected issue: the type of the object identifiers are Integer where those of documents are Long. This is completely abnormal since we have several objects per documents, and therefore we need more distinct ids for objects than documents. I have therefore upgraded the ids of objects to use Long, and provide an implementation that use the lowest 64bits of an MD5 key for object in the same way I do for documents. This implementation introduce two new serializer: UidStringEntityReferenceSerializer and LocalUidStringEntityReferenceSerializer. I have also bridged the statistics that derived from objects. The new implementation works perfectly on a new database but... ... I am unable to provide a proper migration procedure, since hibernate cannot manage changing the types of existing columns. It does not complains during schema update, it simply do nothing about them. And later the data migration breaks since my hashes cannot fit in the database properly. After thorough googling, I understand that hibernate schema updates were not made for production use and are really limited opposed to the general idea we had of it. Since I am currently stucked, I propose to move ahead and use a new tool to properly manage our database schema upgrade: http://www.liquibase.org/ Liquibase is Apache licensed and provide a database agnostic version system for migrating databases. It does not works like a diff tool, but more like a patch tool, where you provide several XML description of your wanted changes and it manage to apply or rollback there changes in an ordered manner to upgrade the database to the latest schema. There is several way to apply these changes and I would like to see if it could be integrated in our current migration procedure, or at least as an independent listener. WDYT ?