David Ketcheson

Sorry, here's a better link: http://nbviewer.ipython.org/urls/dl.dropboxusercontent.com/u/656693/Named%20fields%20as%20a%20dict%20of%20references.ipynb

You're probably right. But just to humor me, could you give an example of something that one might normally do that would create a copy?

Alright, I'm convinced. But let me propose something different. We could implement `solution.fields` as a read-only property. This would guarantee that it's always an up-to-date view of the appropriate part...

Yeah, the fact that the initialization does nothing if there are zero or 3+ arguments is very bad. Looking through the code, `frame` is used quite heavily; there are also...

Edit: and `frame_num`. To be fair, I think this is the result of many different people editing the code.

I think we can do something reasonable in a backward-compatible way: - Add a case for `len(args)==0`, where we check which keywords have been specified - Also add an exception...

I've implemented what I suggested in #636 . Could you both take a look?

I am suggesting revising the directory structure so that coveralls can correctly detect our test coverage. I think that less tricky symlinking would be beneficial anyway.

@weslowrie To be clear, you're concerned that something may be wrong here **only when using a capacity function**, right? Could you point to a specific f-wave Riemann solver that you're...

Could you link to a full example that reproduces this behavior? Thanks.