We are... Spark's DataFrame is essentially a relational algebra AST-builder. Microsoft's LINQ interprets SQL directly in at compile-time. All of these, however, run queries more or less directly in the system in which they're specified.
It helps to think of SQL strings as an untrusted wire format. Yes, parsing is a pain, but it comes with two main benefits:
(i) The wire format is human writable/interpretable, with all the accompanying benefits, and
(ii) The wire format is easily extensible in a predictable way.
That latter one is particularly useful in keeping SQL's ecosystem open. Take a front-end library like SQLAlchemy or ScalikeJDBC for example. It's not practical for any one such library to support every extension provided by every database engine. SQL provides a fall-back for when you need a back-end feature that hasn't been implemented in any given front-end.
Garbage collection has a huge, and generally entirely unappreciated win when it comes to threaded code. As with most things, there are tradeoffs, but every reference counting implementation that I've used has turned any concurrent access to shared memory into a huge bottleneck.
It helps to think of SQL strings as an untrusted wire format. Yes, parsing is a pain, but it comes with two main benefits: (i) The wire format is human writable/interpretable, with all the accompanying benefits, and (ii) The wire format is easily extensible in a predictable way.
That latter one is particularly useful in keeping SQL's ecosystem open. Take a front-end library like SQLAlchemy or ScalikeJDBC for example. It's not practical for any one such library to support every extension provided by every database engine. SQL provides a fall-back for when you need a back-end feature that hasn't been implemented in any given front-end.