Re: Bug #33255 (Windows cannot open files with non-ascii names)

[Ben Pfaff]

On Fri, Oct 09, 2015 at 07:29:12AM +0200, John Darrington wrote:
> On Thu, Oct 08, 2015 at 09:29:35PM -0700, Ben Pfaff wrote:
>      On Thu, Oct 08, 2015 at 06:47:05PM +0200, John Darrington wrote:
>      > In addition to the changes I've submitted for review, I think the 
> resolution to this 
>      > problem involves work in src/data/make-file.c
>      > 
>      > The issue is, this module contains a lot of calls to posix file 
> functions, such as 
>      > open, rename, unlink etc.
>      > 
>      > These work in w32 so long as there are no non-ascii characters in the 
> filenames.
>      > To get other characters to work, we have to do two things:
>      > 
>      > 1. Convert the filenames to UTF-16LE
>      > AND
>      > 2. replace the function calls by their WideChar equivalents.
>      > 
>      > Step 2  however has caveats:
>      > 
>      > open --> _wopen is no problem.
>      > unlink --> _wunlink is no problem.
>      > rename --> _wrename will not work, because windows rename refuses to 
> overwrite an existing file.  We will 
>      >        have to use the native Windows function MoveFileW
>      > 
>      > Not sure about stat.
>      > 
>      > 
>      > It may end up being easier to have a platform specific implementation 
> of make-file.c
>      > 
>      > What do you think?
>      
>      Can this be fixed in Gnulib somehow?
> 
> It probably can be.  The question one has to ask is does it make sense to do 
> so:
> 
> The awkward issue is the _wrename function.   The POSIX rename silently 
> overwrites
> the target.  The Windows rename (and presumably the corresponding _wrename) 
> however 
> fails with an error if the target exists.
> 
> Now some years ago, we (you actually!) contributed a replacement rename to 
> Gnulib 
> such that it behaves per POSIX on Windows.  I think it is perfectly possible 
> to 
> write a replacement _wrename for w32 which behaves in a more POSIX like way,
> however then it will not behave how Microsoft's documentation says.
> 
> Perhaps it might make sense to have a "widechar" version of rename in Gnulib, 
> but
> call it something other than _wrename.

I guess I was assuming that we'd provide a replacement rename(), or I
guess enhance the existing rename() replacement, so that it accepted
UTF-8 file names.  It could be implemented in terms of MoveFileW or
whatever is appropriate.