Refactored _execute to make it more readable (using class instead of closure)

less likely to make incorrect assumptions while reading tracebacks

James added to AdvancedIncSubtensor1 the ability to do set intead of inc. But forgot to update the optimization to pass this new information.

This is a real bug introduced in the trunk but not released.

BUG: str -> basestring isinstance checks.

added a config variable to control the verbosity of exceptions

Add verbosity config

Add test for new fonctionality added by James.

review comments from @delallea
-moved previous developer-oriented description to code comments

Add link in the tutorial to all random distribution.

Make sure the gpu part is only tested when we are actually running on a GPU

Doc sparse

Doc

Make available the list of op that support the r op and the list of r op implementation that is not tested.

Shared pickle

Coding style

Allow Advancedsubtensor1 on gpu to also work on broadcated tensor.

(Thanks fred for writing test)

Add the dtype param to tensor.{zeros,ones}_like as numpy. Test them.