Crutcher wrote:
> Skipping ahead, let me try to rephrase this.
>
> First, this isn't really a python question, it is SQL, XSLT, and
> program design, but I'll try to answer.
Well, first of all, it's about mapping XPath onto a relational data
model. This is clear from the original posting:
<xslt:template match="options_table[field='SERVER']">
<OPTION><xslt:value-of select="name/text()"/></OPTION>
</xslt:template>
Here, it makes sense to consider instances of options_table elements as
rows in the options_table table, just as described. Perhaps, if field
and name are columns in options_table, I would rather model them using
attributes:
<xslt:template match="[EMAIL PROTECTED]'SERVER']">
<OPTION><xslt:value-of select="@name"/></OPTION>
</xslt:template>
Is this possible or wise to do? Well, I did myself write an
XPath-to-SQL query engine, together with a simple "document relational"
mapper in order to be able to execute XPath expressions similar to
those above. However, I don't believe that my code really delivered on
the concept because...
* Generating efficient sets of queries can be tricky: a simple
descent into what appears to be an XML document needs
to involve a lot of joins to enforce the virtual "element
hierarchy".
* One has to decide whether to enforce a schema or not: if you don't
this means that XPath axes like "child" can behave like
"descendant-or-self". If you do, generating "descendant-or-self"
queries becomes very difficult (as far as I can tell and remember).
* The XPath specification states that nodes are returned in document
order - this is difficult to enforce unless you litter your tables
with additional information (again as far as I remember).
* Generating simple DOM structures for simple SQL queries is
straightforward, but generating them for arbitrarily complicated
XPath queries mapped to SQL is quite difficult. If you permit axes
like "ancestor-or-self" and "parent", or have any query that
returns a node over and over again, it's best to instantiate that
node only once in order to keep the number of instantiated
objects low and to more easily test relationships between nodes.
It's all awkward if not particularly difficult.
> You have templates, they contain general layout stuff, and input
> fields. You are transforming them into HTML pages, and part of what you
> want to do is to expand the enumerated fields with the enumeration
> values at the time of template application.
>
> You read an article about XSLT, and decided that you could use it in
> your app. For some unknown reason, perhaps involving large amounts of
> alcohol, you want to use the syntax of XSLT, but have it actually
> parsed by python, which silently does an SQL querry to give you your
> results.
I don't know what it is with comp.lang.python/python-list these days
and the cheap put-downs. Unless you know the person you're responding
to personally, and thus the above counts as some kind of banter, you
would do better to keep the insults to yourself.
> This is a really, really bad idea. There are many ways to solve your
> problem using XSLT, and though all of them are easy, none of them are
> so simple that I could put them in a post to someone who doesn't know
> XSLT. If you know XSLT, follow the directions below, if you don't, DO
> NOT USE XSLT for this project. And for the love of anything you may
> believe in, don't try to implement a partial XSLT engine in python.
Why not? 4XSLT is in Python, or at least large parts of it were.
Moreover, my XPath query experiments used a variant of PyXML's XPath
library which could quite probably have been integrated with 4XSLT,
although I didn't try it at the time.
[Patterns]
Anyway, I'm not suggesting that the XPath-to-SQL route is the right
way. The various patterns you suggest are easier to comprehend and
require a lot less magic, even if there's a bit more integration to be
done to prepare data for use by the templates. But given the existence
of XML database systems, I don't think the idea of accessing them (or
things pretending to be like them) via XML technologies is outrageous
at all.
Paul
--
http://mail.python.org/mailman/listinfo/python-list
