Which would be the less difficult process?
We've been on Confluence since v2 in 2007. We've migrated to v3 and v4. At some point Atlassian changed their schema, database collation to case sensitive and strings to unicode. The upgrade process to v5 falls on it's face when adding a new user_mapping table (expects nvarchar when linking to other tabled). Changing database collation (drop all links and indexes, change collation, switch columns to nvarchar, rebuild all links and indexes) was attempted with unsupported tools but would leave the database in an unsupportable state. Export to XML (in order to import to a database with new schema) misses a required file (https://jira.atlassian.com/browse/CONF-12206). And then the import process fails "java.lang.OutOfMemoryError: GC overhead limit exceeded". 4 weeks burned so far and haven't even started testing plugins on Confluence 5.
Working with Atlassian support Reference: CSP-124664