Re: [Aqsis-development] aqsis source tree rearrangement

[Chris F. <fo...@ph...>]

Hi Paul,

On Wed, Sep 24, 2008 at 01:20:24PM +0100, Paul Gregory wrote:
> In my opinion, this work has already been started to a very small degree
> with the separation of the interface headers into their own directory. If I
> understand properly, what you're proposing is an extension to that, moving
> all the headers into a separate subtree in the source base.

That's pretty much the idea I suppose.  I've always been confused by
which headers belong in the interfaces directory so I tended to avoid it
until now.  I don't much like the fact that headers from disparate
libraries are mixed together in there either - it makes it much harder
to understand which header belogs with which library.

Something to consider is that the headers directory I'm proposing can't
just consist of pure interfaces: the necessity of supporting templates
and inline functions forces us to include some implementation details as
well.  In aqsistex, I tried to put anything which was a true
implementation detail inside the "aqsis::detail" namespace (following
boost).  I probably need to consider making more things details, at
least until aqsistex stabilises somewhat...

> I can't think off the top of my head of any serious implications to
> this approach. I welcome the reorganisation within the project.
> However, we will have to think carefully about what we publish as part
> of an installation, so that there are no problems with dependencies. 

I definitely agree with this - IMO some of our most interesting
interfaces (like the shadervm) are not ready to be exposed to the
outside world, but I'd like to put a framework in place to do so.  In
the meantime, I think the framework will be helpful internally because
of the regular structure it provides.

> In fact, the dependency stuff is perhaps something we should look into
> at the same time, I'm sure there is a lot that could be done to reduce
> the dependencies within our headers.

Yeah, I guess the libraries should have a strict tree-like dependency
structure (ie, no circular dependencies).  I don't think this is quite
true for the core / shaderexecenv / shadervm coupling, which I've always
found difficult to get my head around.

> I'm sure it'll be a big job, but along with some refactoring of our various
> libraries, I agree that Aqsis could then become a more generally useful set
> of libraries, that just happens to have a number simple glue applications in
> 'aqsis'. 'teqser' etc.

Exactly, this is what I'm excited about :-)

Cheers!
~Chris.