That allows us to load that file as a pandas frame.

Full test suite passes.

This fixes #29 

To achieve this goal, I did:
* switch to l2error comparison for all manufactured solution examples
* implement mixed expressions to allow that for systems
* fix `stokes_stress_symdiff`

The following issues remain, but the tests are deactivated until an upstream problem is fixed:
* `heatquation{_dg}_explicit` fails, new issue is #30.
* `poisson_dg_matrix_free` needs a suiting solver, depends on #31 kind of.

I also improved error handling: No more exceptions, instead non-zero error code. That way you get vtk output for failing tests (the previous behaviour was potentially harmful, as you could look at old solutions in debugging!!!).

See merge request !23

Instead return a bool that influences the return code

Switch all poisson examples to comparing l2 error



See merge request !22

Always enable timings



See merge request !21

Add time measure capabilities

Fixes Issue #23.

See merge request !19

Feature/ini options

* Read formcompiler options from ini files
* Have a systemtest macros allowing variation in formcompiler arguments
* Rewrite the test suite using it

Exhibits some more problems in our test suite.

This fixes #1.

See merge request !20

[cleanup] Make function names more consistent

Small name changes. I succesfully ran all tests.

See merge request !18

Add heat equation DG tests

The dirichlet_boundary Expression in the UFL file will be used as initial condition. PDELab figures out that it dosn't need any constraints.

See merge request !17

Use the caching infrastructure for registering function manglers

This fixes #14 

See merge request !16

Feature/cellwise constant

Make it possible to mark Expressions as cellwise const and put evaluation in pdelab outside the quadrature loop. Closes issue #27.

The implementation is kind of sketchy but I wanted to go to weekend. Some notes/TODO:

- I had to add a statement to the ufl file. For cellwise const expressions a dg0 dummy space is used (-> cell wise const), else we use a dg1 space (in execution.py).
- I now use 'default_declaration' from loopy.py for defining parameter objects.

See merge request !14

Feature/modernize cache infrastructure

* use local storage instead of global one
* Remove cache context
* Remove cache item types
* Introduce cache inspection method

See merge request !15

They are concepts that are not really necessary with local
storage, that I do not really want to explain to other people.

Only register generators globally. This makes inspection of cache
items much easier.

Feature/evaluate basis gradient with loopy 2

Second try to implement issue #22. We avoid problems with diagonal matrices in the following way:

1. The user can give the option 'diagonal-transformation-matrix' to the form compiler if he wants to reduce the loop.

2. It is now possible to set options during the form compilation. When a YaspGrid is created we automatically set the option 'diagonal-transformation-matrix'.

Usually it makes little sense to provide the information yourself. I just wanted to add the possiblility if someone has a axiparallel grid that cannot be a Yasp Grid (eg. due to non rectangular domain).

Closes issue #22.

See merge request !13

Cleanup/default no dot files

New default: Write no dot files.

See merge request !12

Feature/instationary

Make it possible to solve instationary problems by providing a form called 'mass' in the UFL file. Some notes:

- The most important change is in cache.py: Deleting the cache did not work. Consider the following minimal example:
```python
    global_dict = {}

    class ModDict(dict):
        def __init__(self, tags):
            dict.__init__(self)
            self.tags = tags

        def __getitem__(self, i):
            # if i == "True":
            #     return True
            return i in self.tags


    def test():
        print eval("True")

        print eval("True", {})

        print global_dict.items()
        print eval("True", global_dict)

        print eval("True", ModDict('test'))

        print eval("not True", ModDict('test'))

        print ModDict('test')["True"]

        print ModDict('test')["Blubber"]

        # print global_dict["True"] -> key error


    test()
```
The modified dictionary never throws a key error. Instead it always returns False if a key is not found in the tags. Therefore `print eval("True", ModDict('test'))` returns False!
- You can switch between implicit and explicit through formcompiler arguments.
- The time loop in the driver is hard coded. Should be replaced with something smarter in the future. -> Issue
-  In order to test the problems I compare with an exact solution. Easiest way to do that: Chose g and f like in the poisson example. Unfortunately this leads to boring vtk files since everything ist stationary ;).
- The explicit time stepping has some weird behavior. The reference solution behaves exactly the same and it is not surprising that an explicit time stepping method is not suitable for soling heat equation. Nonetheless I think it is strange how it behaves on the boundary (it doesn't try to satisfy the Dirichlet condition). The cause for that could be:
    1. I'm doing the wrong thing in PDELab.
    2. You shouldn't do explicit instationary CG in PDELab.
    3. Some bug.
- Probably it is the first option ;).


See merge request !11

This reverts commit e1bbce4c.

for style reasons