On Freitag, 29. Mai 2009, Jan Janak wrote:
> [..]

Hi Jan,

sorry for the late reply, i was on vacation.

> > > What about importing them into the tools directly?
> >
> > do you mean the tools directory? Hm, not sure what then the purpose of
> > the scripts directory would be, perhaps this could be then removed?
>
> Currently the script directory contains all the database related script,
> and some stuff that is not being used much (or at all), i.e. filter_log.sh,
> serstats, kam_to_sr.sh.
>
> I think we can use the scripts directory to store supporting scripts that
> people typically run once or occasionally.
>
> The tools directory could than be used for all other admin related tools,
> such as sercmd, gen_ha1, kamdbctl.
>
> We can also get rid of the utils directory.

Ok, sounds good.

> > > > The xml definition of db tables were in db directory, which now is
> > > > library srdb1(same for ser, are now in srdb2). Do we let them
> > > > there or is another more suitable place for them as they are not
> > > > dependent of
> >
> > DB
> >
> > > > api but of what DB structure to be used (some modules using db
> > > > were moved in common folder already)?
> > >
> > > I would suggest that we keep them there for now. The scripts that
> > > are
> >
> > used
> >
> > > to generate db schema need, in my opinion, an overhaul, so we might
> > > just
> >
> > do
> >
> > > it in one go.
> >
> > I refactored the k XSL scripts (which were imported from SER) in the
> > last year somewhat and also improved/ added support for DBs!=mysql. They
> > do IMHO a good job now to generate the needed database scripts from XML
> > sources, so i don't see a big need to rework them. What i removed was the
> > possibility to also create insert statements, i think, not sure if this
> > is needed from SER.
>
> I think that xsltproc (or XSLT) is the wrong tool for the job. If you take
> a look at the conversion scripts, they use lots of variables, conditional
> statements, recursions, and functions specified in the XPath specification.
> In other words, we use xlst as a real programming language, but unlike many
> other languages available today, it has horrible syntax. Heck, even Perl
> scripts are more readable than XSLT.

Hm, i agree that the xpath stuff is really a bit ugly. :-)

> [..]
> Another major problem is that the scripts are not portable. They work with
> xsltproc but would not work with other xsl processors due to subtle
> differences in implementation. I couldn't even generate a consistent output
> across different versions of xsltproc.

Only used xsltproc so far, but i also got the impression that portability accross different parsers is sometimes a problem.

> If I had to write such scripts again, I would have chosen another language,
> probably Python or Perl. They are both wide-spread, they have decent XML
> parsers and less terryfing syntax (well, in case of perl this is
> questionable
>
> :-).

:-)

> With python or perl we could do much more than just generate plain-text
> output, we could create the tables directly in the database, setup
> usernames and passwords, handle backups and upgrades conveniently in one
> tool, and so on.
>
> While writing those scripts in XSLT was a good experience, I think we are
> using the language outside its domain. But at least we learned it, which is
> good if one needs to modify the docbook xslt scripts, for example.

I don't would write this stuff against in XSTL either, learning all this was a interesting experience.. ;-) I only wanted to say that i don't favor a complete rewrite of this stuff _with the same functionality_ just because it turned out to be not the perfect choice for the job. The existing implementation works so far ok, and IMHO there are other, more important problems right now. If we get more functionality the way i have nothing against it at all.

Henning