One of the problems with learning a web framework is that there are a lot of them, and the effort it takes to find out which one best serves your needs can seem to be on par with rolling your own, without the assurance that you'll actually get your needs met. And I suppose the ORM issue faces the same decision-making problem.
Hand-rolled crap is at least crap that was rolled to suit the problem you're facing, and you'll do it using tools you understand. You're not learning "any" framework, you're learning a bit of them all and hoping that's enough to point you in the right direction. Sometimes it's not, and then you're stuck with a hammer for screws and someone else's code that you don't understand.
I see learning a framework as less about trying to get all the things you need, and more about 'borrowing' someone elses architectural thinking. When you encounter something the framework doesn't provide, you probably have all the architectural tools available to you to create something good that fits in well with the rest of the framework. Also, when you need to move to another framework, you have a base of experience you can draw from to help you learn much faster.
Hand-rolled crap is at least crap that was rolled to suit the problem you're facing, and you'll do it using tools you understand. You're not learning "any" framework, you're learning a bit of them all and hoping that's enough to point you in the right direction. Sometimes it's not, and then you're stuck with a hammer for screws and someone else's code that you don't understand.