An ORM will force the class structure onto the RDB-schema.

It's only using a subset of SQL and can be mapped back 1-to-1 to an object model (of the particular host language, here Perl).

It's like a amputated creole, speaking English words with Latin grammar. But without the possibility to introduce new Latin particles to express the new grammar, forcing you to use costly workarounds.°

E.G. there is no "inheritance" in SQL.

An SQL-Abstraction attempts to express all or most of the SQL-Space inside the host language, like with a query builder.

It's effectively just a new SQL dialect, attempting to map 1-to-1 onto other dialects (like LIMIT vs TOP )

My impression is/was that DBIC attempts to have the best of both worlds, which also explains it's complexity.

For a long read why many people despise ORMs you may want to have a look at

Object/Relational Mapping is the Vietnam of Computer Science

(I'll skip the first 6 pages about US-Vietnam-War though. :)

This blog-entry had a heavy impact and was often discussed in both directions, you might want to google the responses...

update

swapped "Chinese" with "Latin" for claritas in metaphora ;-)

°) or is it rather speaking Latin with English grammar? :)