I'd say there are problems on 3 levels: 1. All the little things, inconsistencies and annoyances. You know, NULL behaviour, identifier usage, things that you j

I'd say there are problems on 3 levels: 1. All the little things, inconsistencies and annoyances. You know, NULL behaviour, identifier usage, things that you just have to memorise (everything is non-orthogonal). 2. Language-level. No real abstraction, no code reuse, weird sublanguages for trivial stuff, no libraries/modules/packages. 3. Standard-level. I mean, this is not standard anybody can use. It is unreadable! I mean I can go a very long way if necessary - used to reading the original C++ and C stds. But ANSI SQL... A bottomless pit. Databases do implement subsets of something ANSI-ish but these vary a lot. All of this results in templated queries, 1000s of lines of queries that are only slightly different and no chance for a truly db-agnostic language. I mean... Have you seen write-once 3000 LOC queries? This is what I can "fall apart"! I like the original Codd's relational alebra, even love it. It is both simple and practical. Most databases use RelAlg-like internal representations that optimisers work on for a reason! But we're stuck with SQL and it will never go away.