You suggest using a DBD::mysql DSN for db:mysql, then? How are they different? I've heard no complaints so far, and the sqitch change was relatively straightforward. Although I guess since it always uses DBD::MariaDB (even with older versions of URI::db), it knows to no longer use the mysql_ options. Perhaps that's what you mean?

DBD::MariaDB is not a drop-in replacement for DBD::mysql, so this change isn't backward-compatible.

I downgraded sqitch to work around the issue I reported over there (the one about the mysql.users table), and ran into even more problems because that version of sqitch was passing mysql_enable_utf8 to the driver, but URI::db was telling it to use MariaDB for a mysql URI.

I suppose at the core, the issue is that a URI::db object can give you a DSN with a driver in it, but the code that asks for that info doesn't actually know what driver that will be; but different drivers support different options, so the code is still going to have to decide what to pass to connect. dbi_dsn is therefore most likely to be used as a helper method—i.e. to make it simpler to pass options to connect—than a decider method, i.e. to pick what driver to use in the first place.

sqitch itself is an example of that: back in version 1.3.0, it was assuming that URI::db was going to tell it to use the mysql driver, and therefore passed mysql options to it. That seems like a reasonable assumption to have made, so any code working on the same assumptions is going to break.