[Hibernate-devel] code generator

[Brad C. <bra...@wo...>]

i have just commited the first (very rough) cut of workingmouse's code
generator to cvs.  over the next few days i will try to make time for
writing javadoc comments.  there is an ant task (codegen) that runs the
code generator over a test mapping file
(cirrus/hibernate/tools/codegen/test/hbmapping.xml) and places the
generated files under a "generated" directory.

mapping file requirements:
all properties and id's in a mapping file must have the type attribute
specified.  currently a type must be a fully qualified java class, with
the one exception being the hibernate binary type.

supported mapping file functionality:
subclasses
nested collections
one-to-many
many-to-one

high-level architecture:
our goal is to use the code generator to generate the javabean classes
defined in the mapping file, as well as many of the other classes, web
pages, etc. that use them.  before writing this i weighed up the idea of
using XSLT.  firstly, i haven't written any XSL, XPATH, X..., for about
a year now, so i am a bit rusty with it all.  secondly, i didn't think
it would elegantly achieve our goal compared to just using java and/or a
template mechanism like velocity.

ok, so at its core, how does it work?  essentially, the mapping file is
parsed into an internal structure and then a renderer used to write the
internal structure into a series of files.  i have included one renderer
(the BasicRenderer class) that simply writes out the classes specified
in the mapping file.  anybody can write more renderers by implementing
the Renderer interface and configuring the code generator to use their
renderer(s).

gavin i did have a renderer lying around here somewhere that used
velocity, which although not as powerful was very easy to use.  i would
like to resurrect it and commit it as well.  however this will introduce
a dependancy to velocity (http://jakarta.apache.org/velocity/).  is that
a problem?

configuration:
the code generator can be used without any configuration, but the
location of the mapping file to process.  this is how the ant codegen
task is setup.  in this scenario, the BasicRenderer is used to write the
classes defined in the mapping file under a directory called
"generated".  however, a configuration xml file can be used by
specifying its location with the "--config=3D" command line parameter.
(there is a test config file located in
cirrus/hibernate/tools/codegen/test)  the config file enables u to
specify generators.  a generator is simply a renderer plus information
about the base directory to write the generated files to, package to put
them in and how to name them.

database reverse engineering:
currently the code generator revolves around the mapping file, because
that is the way we like to work.  however, i can't c any reason y the
parser code couldn't be abstracted away behind some sort of Parser
interface and made configurable.  this could enable a "database parser"
that uses the JDBC meta data of an existing database to build the
internal class structure.  the problem would be in writing the single
mapping file, as the architecture isn't really geared for that.  i am
sure we could come up with something though.

hope this is useful for people.

brad

> _______________________________
> brad clow
> chief technical officer
> workingmouse
>=20
> email: bra...@wo...
> web:   http://www.workingmouse.com
>=20