[Zope3-dev] proposal to extend meta.zcml files with documentation strings

[R. David Murray]

On Mon, 23 Sep 2002, Shane Hathaway wrote:
> R. David Murray wrote:
> The main trouble I've had with ZCML is that it's hard to find the Python
> code it supposedly invokes.  In fact, it seems to try to hide the
> association.  I realize there need to be multiple passes over the
> configuration (Jim and I came up with the multiple pass idea together,
> after all!) but the current coupling between directives and executable
> code is much too loose.  Figuring out exactly what code will be executed
> as the result of an active directive is too difficult.

There might be a way to use the parser to at least build a map
between the directives and the generated actions.  I'll look into
that if I get the time.

> should be represented as an object.  A configuration file should be an
> object.  An active configuration should be an object.  I think it could
> lift a lot of the complexity of Zope 3 configuration.  I've been
> drafting rough interfaces, if anyone is interested.

This sounds very reminicent of a long thought dump Philip Eby posted
to the transwarp mailing list a couple months or so ago.  If you didn't
see it, Maybe you and he should talk <grin>.

> Also, I think configuration in XML has great benefits, but coding
> "meta-configuration" in XML adds to the complexity burden without a
> benefit that I'm aware of.  We envision site developers changing the
> software configuration, but do we envision them changing the software
> configuration configuration?  Imagine you're designing a home through a
> contractor.  You'd want to specify the layout of the rooms, but you
> wouldn't want to specify the format of the blueprints.  That's up to the
> contractor.

This is true.  However, I will note that I built my prototype
doc-generator by reprograming the meta configuration directives.
The "default" zcml binds routines to the meta-configuration
directives that simply ignore the docstrings.  A special
"docgenerator" zcml file, used only by the doc generator
program (that I have yet to write <grin>) binds the directives
to routines that do something useful with the doc strings.
This may be a very specialized use case, but it seems to me
that in principle it might be useful for a product developer
to be able to modify the configuration directives of another
package without touching that package, just like we can now
modify the actual directives without touching the other
package.

Phil's system sounded pretty sweet, though, and I imagine yours
would share many of the nice characteristics.  I certainly
wouldn't mind doing away with zcml metaconfiguration myself.

On the final hand, I've got this thing 70% working, so it sounds
like it's probably worth deploying until/if something is done to
improve the basic design of the configuration system.  I did some,
um...small underhanded things in meta.py that I'd best check out
with Jim, though <grin>.

--RDM