Why would bringing a framework in house be better than using open source?

I'm confused by all these people that thing 'Rails is crap, clearly people should roll their own everything since that will be more reliable and better engineered'

If you think Rails is crap, the only answer is to look to 'more mature' platforms, not roll your own. Spring/Hibernate might be a reasonable answer. Django is not, being at a very similar level of maturity. Nothing in PHP even begins to approach the level of maturity.

Software engineering is hard. Really hard.

No software is a panacea. Decisions are made which favor some use cases over others.

When you are as large as google or facebook or github, those decisions which don't naturally mesh with your workflow become more painful over time. And when you have enough in-house resources, the cost to working with the existing solution far exceeds that of extending it on your own or building your own.

Hiphop is a good example where Facebook saw certain things they wanted to optimize in PHP, and it made far more sense for them to work on their own thing (many of those decisions underlying hiphop make sense for facebook but not for most companies deploying PHP)

I sometimes wonder if a good alternative to django/rails is to use the most minimal framework that's reasonable to start with, and then gradually customize it away.

This actually sounds like a promising idea, though only if you've got insane dev chops on the founding team. In the Ruby world, Sinatra would be the obvious place to start. On the other hand, you might just end up reinventing Rails.

I love Sinatra, but there's a long-running joke that if you look at the source for any Sinatra application with even a little complexity, you'll find a half-baked re-implementation of Rails inside. I've built apps like that and it's definitely not an exaggeration.

Ah, so following Greenspun's tenth rule [1], perhaps we have this:

  Any sufficiently complicated web application contains 
  an ad hoc, informally specified, bug-ridden, slow 
  implementation of half of Ruby on Rails.

[1] http://en.wikipedia.org/wiki/Greenspun%27s_tenth_rule

That was my road to Ruby: Developing in PHP "mvc" frameworks. Even had a PHP active record implementation. Create function was used heavily ...

http://php.net/manual/en/function.create-function.php

That's probably almost the exact quote I saw :)

Yep, especially as there are sinatra stacks that already do that for you in a sane fashion, e.g. Padrino.

Also: yes, building from minimal makes you appreciate the work that big frameworks have already done.

Tried it. You spend way too much time solving problems that are already easily solved in the existing larger ecosystems.

Yes indeed, and also leaves everyone in the muck when the lead dev leaves. Your left with a load of code you had wish you had documented as well as the previous frameworks. The value of existing frameworks is the documentation and stack overflow answers.