[Gneuralnetwork] Adding a build target?

[Ray Dillinger]

I have written, tested, and debugged a parser and writeback utility for
(the basic part of) the new network representation.  I would like to
make a second build target now for a version of the program that uses
the new representation, in order to facilitate testing and further
development, because that would greatly simplify and disambiguate how
and where to add the new code to the project.

I've been pulling routines out into a separate project to write and test
code then sticking the routines back into the files to check it in, but
this does not create a configuration where people can run and test the
new code without pulling it out and assembling it into an independent
project themselves, and that's crazy. It's time for that to stop.

But I've only the most basic acquaintance with automake, and the
makefiles it generates appear to me to be ridiculously complex and
fragile piles of gibberish that depend on way too many moving parts and
additional code to be robust or understandable.  Normally, I would
simply add a file with a second 'main' routine, and a one-line addition
to the makefile to add a second build target.  But just figuring out HOW
to get automake to do this looks like a project in itself.  I could
splice it in as something different that the main executable does if
tickled with the right command line arguments, but is that what people want?

More to the point do we really need this 800-pound gorilla in our build
process?  Is it worth my time to climb that hill and figure out how to
get automake to cooperate and configure what would normally be a
one-line change??

The new representation - now with parser, writer, and feedforward - is
ready to be its own executable.  But it doesn't integrate well with the
existing code, and I'm not sure in what form I should check it in.


                                Bear

signature.asc
Description: OpenPGP digital signature