to ensure that any code looking for it
(e.g. jenkins configuration files)
will still work.

if the user does explicitely set the cxx config flag in THEANO_FLAGS.

to make theano scripts installed when running pip install.

These step are done to fix issue #5308. The bin folder
containing the scripts is now importable from Python,
with each script embedding its code into a main() function
specified as the entry point of the script in `setup.py`.

We also detect a Windows-specific error: the cache folder
cannot be deleted because it contains shared libraries that are
always loaded at theano importation and stay opened on Windows,
which forbide the deletion of the shared library files in the cache.

The workaround chosen is to unset theano config flag `cxx` only
on Windows platform and only in theano-cache script. Doing that,
Theano will not try to compile any C code, and will also not try
to load compiled C codes that are in the cache, allowing these files
to be deleted.

To ensure this workaround does work, we also ensure that lazylinker
module will be also ignored from cache when `cxx` is unset. Previously,
if cxx was unset, Theano would not compile lazylinker but would look
for an already compiled version of this module in the cache.

[ENH] Multinomial without replacement

compatible when axis < 0 in tensor.roll(x, shift, axis=None)

Complete Add grad_scale Op

Add a new grad_scale op to scale or inverse gradient in backpropagation.

uint16 added into _good_broadcast_binary_normal

theano.pp crash fix and display fix

Get rid of redundant copy for GPU "map" style scan

Add regression test for nvcc.compiler_bindir (#5373)

Bug: gradient of ifelse with gpu inputs

Fix the buildbot.

Mixed fixes

gpuarray: Revert part of c6ffa460

Select the dnn convolution algorithm using actually available memory

crash fix. Add the missing inputs to the function call. It also need the real data. I don't sync as we don't care of the values.