Re: [xslt] create Directory bug in libxslt for W32

["Aleksey Gurtovoy" <agurtovoy meta-comm com>]

Daniel Veillard writes:
> On Tue, Nov 30, 2004 at 03:10:38PM -0600, Aleksey Gurtovoy wrote:
> > Thomas Fischer writes:
> > > Hi folks,
> > > i found a bug in libxslt and Daniel Veillard told me, that's
> > > a win only bug and he can't help   :o<
> > > 
> > > The Bug occurs, when i try to create a directory via command
> > > line parameter --output (like "xsltproc -o does_not_exist/res ...")
> > > or via <exslt:document href="does_not_exist/res"
> > > 
> > > Anybody knows a workaround or patch for this problem?
> > 
> > See the attachments for the latter (against the current CVS).
> 
>   I do not understand the first patch nor the associated
> side effects. On windows the file path separator is '\' why is this
> wrong and need to be removed.

I agree that the patch is a little controversial, but only due to the fact 
that 'xmlParserGetDirectory' is used to extract directory paths from both 
URI/URLs _and_ file system paths. For the former the running platform is 
irrelevant; for the latter it's not (obviously). I'm not that familiar with
the libxslt/libxml2 codebase to be able to say in how many cases when a
function is passed a 'filename' argument it actually _is_ a native 
filesystem path (as opposite to a URL), but I assume that at least some 
of them are. Of course you're in much better position to answer this.

In any case, my take on the issue is that, ideally, all current calls
to 'xmlParserGetDirectory' should become calls to either 
'xmlParserGetURIDirectory' or 'xmlParserGetFileDirectory' (both currently 
non-existent), the latter differing from the former in that it would 
respect the running platform's conventions.

Does it make sense to you?

>   Moving the stat() remapping to the place where it is used makes sense
> to me, so that one will be applied, thanks,

You're welcome!

--
Aleksey Gurtovoy
MetaCommunications Engineering