ast,runner: Remove template field from connector

This allows us to more easily extend the mechanisms for constructing
components and connectors.

trivial: Style unstyled files

Remove makedirp

Dead code removal.

camkes: Use custom Jinja loader for tracking files

We can override the FileSystemLoader that Jinja2 provides to track which
template files are opened during the rendering process. This makes it
easier to generate gcc-like dependency files.

camkes: Remove version() function

This removes the functionality that hashes the camkes sources.

runner: remove vestigial options and workarounds for template cache

runner: Use RenderState to capture persistent data

restore 'outfile' context as 'outfile_name'

Isabelle theory templates need to know the output filename. Hence, we
provide the filename as 'outfile_name' when running non-code templates.

Only pass required options to renderer

Passing only the required options when instantiating the Renderer ensures that the Renderer
is not depending upon unexpected or specific options

Fix licenses

Revert "runner: Configure jinja to remove some whitespace"

This reverts commit f05595dae4f0356044442d427a0b20b513d9d053.

runner: Configure jinja to remove some whitespace

Add interface for adding fill information to symbols

capDL frame objects can have an optional 'fill' attribute that instructs the loader
with what to initialize the frame with when loading. This change adds an interface
for adding such information to an existing page sized/aligned symbol

Add interface for specifying symbols to keep

Added `lookup_template` to template context

This allows templates to determine the existance of other templates
as they are rendered. This is required for templates to determine
whether connectors have header file templates.

Output of --version reads from VERSION file

This adds a function "version_number" for reading the version number
from the VERSION file, and renames the existing "version" function
to "version_hash" to disambiguate between the two.

Update import usage within the runner for Python changes.

The aim of this commit is compatibility with Python 2.7.11. These changes
effectively make the runner a first class module (relative imports used
internally; __main__.py importing sources as belonging to a module). The intent
is to remain backwards compatible down to Python 2.6 and forwards to Python
3.5.

Refactored parser init commit.

Use `mkdirp` in preference to `os.makedirs`.

When running multiple instances of the CAmkES code generator, a race condition
is sometimes provoked whereby a call to `os.makedirs`, after checking the
target directory does not exist, can fail because another instance of the code
generator created the directory in the meantime. This change makes the code
generator robust against such effects, allowing multiple instances of it to
execute in parallel.

Optimisation: Disable auto-reloading of templates.

This (default) feature of Jinja is designed for an environment like a
webserver, where templates are expected to change on disk while the service is
still running. CAmkES' code generator does not work as a persistent service and
templates are not expected to change while the generator is running. The net
effect of this should be a (negligible) performance improvement.

Resolve symbolic links in template paths.

Pass the entire assembly to the template context and allow templates to look through both the configuration and the composition of the assembly

Optimisation: pre-compile templates when the compilation cache is enabled.

This commit makes some changes that kick in when the compilation cache is
enabled. If it is the first time we have seen the templates, they are compiled
into Python byte code and stored in the compilation cache. When rendering
templates after this, we can just retrieve the pre-compiled version from the
compilation cache.

It is difficult to measure the effect of this because it is optimising for a
case when you have edited your ADL (cache miss on the rendered template itself)
but have not modified CAmkES itself. A contrived benchmark gives an optimistic
performance improvement of 5%, but too much weight should not be given to this
figure.

Release snapshot