> On the FormatterWalker - I have to ask a stupid question (:<). I can
> base it on the XercesTreeWalker class, but I need
> to implement the various FormatterListener start/create calls. The
> problematic one is startElement as I need to wrap the
> DOM_NamedNodeMap in a wrapper that implements the SAX parser
> AttributeList. Does such a thing exist, or do I need to create a new
> XercesDeprecatedNamedNodeMapAttributeList (the names just keep getting
> better) and then use a string pool to map the old DOMString objects to
> XMLCh?
Yes, you'd have to create something like that, so maybe we should just skip
it. I hate to add more and more code that isn't likely to be used. In
fact, Xalan has really grown lately, and I'd say we're due for some spring
cleaning...
> (Another) question for you. The issue with -fno-elide-constructors
> raises an interesting question in regards to compilers and makefiles.
> Is there any scope for testing these kinds of things in the configure
> script? That's the real power of autoconf, but I notice the preference
> seems to have been to go down the path of options in various header
> files. I've played a bit with autoconf in the past (I use it heavily in
> the build process for xml-security), and if we can do a programatic test
> for it (which could just be a matter of trying to compile something with
> the STL in it), we can build the option into the autoconf script.
>
> The downside is it complicates the configure script and it means that
> whenever _anything_ changes in the OS, you need to re-run configure. At
> the moment, you simply re-run make and it works everything out
> automatically.
>
> The up-side is it simplifies the makefiles and the header files and
> means that when new compilers and versions come along, there is a good
> chance the configure script can "work out" the correct options for the
> given compiler.
I'm all for using autoconf more where we can, but since it doesn't work on
Windows, we can't really get rid of all of the configuration stuff in the
header files. That said, things like no-elide-constructors are perfect for
autoconf, because we can detect the version of gcc being used and behave
accordingly. In fact, I'd say our Makefile could use a general overhaul --
It's been hacked and ifdefed to death.
Dave
