Re: [xml] Problems with file names in UTF-8 on Windows

[Daniel Veillard <veillard redhat com>]

On Thu, Aug 31, 2006 at 08:53:08PM +0400, Emelyanov Alexey wrote:
> > > Here is the reworked patch...
> > > What has changed. It operates now in "paranoid" mode...
> > > [...]
> > > Roland
> >
> > Okay, this seems to answer all points raised in the past, and is
> > changing windows only code (except for the duplicate function removal),
> > so I applied it and commited it to CVS.
> >
> > Daniel
> Hi.
>
> After updating the library works under Win 9x. Thank you.
> Although code has become rather unclear...
>
> Besides some problems are possible.
> First, latest Microsoft compilers not use msvcrt.dll any more, and link
> programs with msvcr70.dll, msvcr71.dll, msvcr80.dll etc.
> Second, the dynamic loading will be used in the
> static version of library too.
>
> Probably, an alternative version of patch will be more acceptable.
>
> Patched library was compiled with MSVC 5.0, MSVC 7.1, Mingw (gcc 3.4.2)
> and was tested under Win 98, Win 2000 SP4, Win XP SP2.
>
> Changes in comparison with version 2.6.26:
> 1. Platform specific code is moved to separate functions (opening
>    of files and status information retrieving).
> 2. As xmlInitParser() should be called in multithreaded programs
>    before use of any other library functions, detection of
>    platform and appropriate API features is carried out in
>    xmlRegisterDefaultInputCallbacks() and
>    xmlRegisterDefaultOutputCallbacks().
>    Thus there is no need to complicate a code,
>    protecting it with the mutex.
>    After initialization two static function pointers are
>    used only for reading, therefore problems with
>    multithreading will not arise too.
> 3. In function __xmlIOWin32UTF8ToWChar() flag MB_ERR_INVALID_CHARS
>    is used for fast detection of filename native encoding.
>    Also unnecessary increase of string buffer length is removed.
> 4. The dynamic loading of msvcrt.dll is not used at all.
>    The functions _wstat()/_wfopen() exists even in
>    msvcrt20.dll (Win 95 distribution).
> 5. Functions xmlMalloc()/xmlFree() are used instead of malloc()/free().
> 6. In a code the stat() calls are used. If there is _stat()
>    function in C library, appropriate macros is defined.
> 7. Code is changed for Windows only.

  Roland's patch has been applied to CVS. So any further patch should
be against CVs head i.e. the version with Roland's code changes. This
would also help checking what you are really changing, please resubmit.
BTW Roland, if you could also check those (or others one Windows) 
I would feel more confident taking a decision as I just can't test those
patches.

Daniel

-- 
Red Hat Virtualization group http://redhat.com/virtualization/
Daniel Veillard      | virtualization library  http://libvirt.org/
veillard redhat com  | libxml GNOME XML XSLT toolkit  http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine  http://rpmfind.net/