You always end up having to do the thing you were trying to avoid plus deal with the middleman abstraction you introduced trying to avoid it. Anytime I hear someone say "- so you don't have to -" I wince and anticipate upcoming technical debt.
So use an ORM so you don't have to write and maintain SQL is bad, because writing and maintaining SQL is not expensive, but using an ORM is (at least in my experience: I've run into many cases where the ORM generates pathological queries that aren't fixable without throwing away all the code that was calling into the ORM; where a bad query in plain SQL is usually readily accessible).
On the other hand, use Erlang's hot code loading so you don't have to restart nodes to update code is not bad: hot code loading isn't super expensive (you have to do a little bit of planning for some types of changes), but restarting nodes is (first you have to move all the traffic, then you have to actually restart it, then you have to move the traffic back).
Not kwowing how to write proper SQL is a problem, and ORM won't solve that for you.
The correct answer is to forget about database independence and double down on just one so you can use it to its full potential.
: And obviously it should be Postgres...
when you are not writing a SaaS, but shipping a piece of software for a customer to run on their own server, it's a very common use case to support different database vendors - especially in enterprise software.
There are non-abstractable differences once you no longer have a toy project, e.g. locking and isolation semantics.
There might be some valid reason for making your applications talk to different DBMS, but you have a tough slog ahead no matter how you do it.
Yeah, sure, go tell that to your manager who promised multi vendor compat to your clients.
Problems with the ORM nowadays always stem from developers who are ignorant of SQL. Those people should not be working for you anyway.
Okay, I'll tell him that 5x the systems means more work, possible bugs, etc. Seems logical.
I'll tell him the same thing if he promises Windows, Mac, Linux, Solaris, and Plan 9 compatibility.
I'm not saying you shouldn't be able to use the same application with several databases; I'm saying you should avoid it.
The working approach is: Use Hibernate (or any other ORM) so that you only need to write the SQL that matters.
There is nothing wrong with using a tool to make your life easier, and there is a lot of necessary-but-borderline-boilerplate work that needs to be done when interacting with a database: validation, caching, transaction management, etc.
An ORM can handle the 90% use case for these things, freeing the programmer to focus on the 10% where the business value actually lives.
Using an advanced tool doesn't remove the need to understand what is going on under the hood: it just eliminates the need to care about the inner workings until they interfere with solving the problem-at-hand.
I'd say focus on the 10% that hibernate sucks at, which will be independent of the business value.
"Use Hibernate", for example, equates to "let third party software I don't understand fiddle around with my bytecode after compilation". That not only crosses the "Nope-line" for me, that even runs past the horizon behind it.
Ignoring that, if a tool like Hibernate requires you to write your business data definitions (i.e. classes) in their specific way, I'd say something's fishy. You'll never be able to untangle the two. Note that, for all their repetitiveness, that's something SQL-based DB interaction approaches would never do.
Edit: It's perhaps noteworthy to mention Uncle Bob's talk about the "Clean Architecture". I think his arguments on how to structure a Codebase based on what is actually important have much much merit to them.
I think the real anti-pattern that leads to ORMs is the fact that languages have no way of natively dealing relationally with data. This is the real origin of the impedance mismatch problem. Languages still deal with data using hierarchical and linear models.
Everything in engineering is about trade offs. You bullshit detector should go off whenever someone acts like there are only pros and no cons. The phrase "so you don't have to" seems to pop up when someone is trying to pull to wool over your eyes.
"So you don't have to" is entirely valid.
You have running water so you don't have to walk half a mile to some well with a bucket. The trade off is the whole plumbing infrastructure and the trade which maintains it.
Not saying the situations are that closely analogous, but a huge part of high level languages is not worrying about that issue.