Re: [Docutils-develop] Trying to fix the stylesheet/stylesheet_path mess

[David G. <go...@py...>]

[Felix Wiemann]
 > The overriding mechanism for stylesheet and stylesheet_path is
 > broken, because if 'stylesheet_path' is specified at any place,
 > setting 'stylesheet' has no effect.

This was discussed here in July.  See
<http://thread.gmane.org/gmane.text.docutils.devel/2068>.

 > So, as you can see from the last command, when specifying
 > --stylesheet at the command line, the stylesheet-path in
 > docutils.conf isn't overriden.  That's bad and unintuitive.
 >
 > There are several fixes I can think of:
 >
 > 1. Delete ``stylesheet`` when ``stylesheet_path`` is set and vice
 >    versa.

That's the conclusion Beni & I reached as well.  I don't think fix 2
or 3 are worth exploring.

 > I tried to implement fix number 1, and encountered the following
 > difficulties:
 >
 > * There is duplicate logic in the html4css1 and the latex2e writer
 >   concerning stylesheets.  I'm not arguing for moving the stylesheet
 >   handling to the Docutils core, but I think that the logic
 >   shouldn't be redundantly stored in two files.  Maybe have a common
 >   baseclass or so?

They do have a common base class: docutils.writers.Writer.  If the
code is duplicated between the two writers, it could be moved to the
base class.  If it's not desirable to have the shared code in
docutils.writers.Writer, we could make a new subclass of that, perhaps
WriterWithStylesheet.

But looking at docs/user/config.txt and at the command-line option
descriptions, the two writers seem to have different semantics.  I
don't really understand the descriptions for the LaTeX writer, and I
suspect that there might be some confusion there.

 > * Setting options is done differently when doing command line
 >   parsing and when overlaying settings settings_spec and
 >   settings_overrides.  So I had to fix it twice.

I don't think settings_overrides etc. should have any special support.
If "stylesheet" and "stylesheet_path" are mutually-overriding options,
they could also become mutually-exclusive settings: if both are set,
an error should be raised.  If client code sets both programmatically,
that's an error.

 > The following output from ``cvs diff`` is an example implementation,
 > but only for the html4css1 writer.  It's extremely hackish and I'm
 > not sure if it will always work.
 >
 > --------------------------------------------------------------------
 > Index: docutils/frontend.py
...
 > +    ATTRS = optparse.Option.ATTRS + ['overrides']

I like this; the "validator" keyword should be implemented similarly.

 > +    def set_defaults(self, **k):
 > +        """
 > +        Hack to make stylesheet override stylesheet_path and v.v.
 > +        """
 > +        s = 'stylesheet'
 > +        sp = 'stylesheet_path'
 > +        if k.has_key(s) and k.has_key(sp):
 > +            # arbitrary
 > +            k[s] = k[sp]
 > +        if k.has_key(s):
 > +            self.set_default(sp, None)
 > +        if k.has_key(sp):
 > +            self.set_default(s, None)
 > +        optparse.OptionParser.set_defaults(self, **k)

I don't think this is useful.  If either stylesheet or stylesheet_path
is set in code, the other should be explicitly unset (set to None).
The code that uses stylesheet/stylesheet_path should check for the
collision.

-- 
David Goodger <http://python.net/~goodger>