Re: [Caml-list] Looking for a windows ocaml UTF-16 encoded filename aware library

[Bob Atkey <bob.atkey@gmail.com>]

Hi Matthieu,

I wrote a little C binding to do pretty much what you are asking:

   https://github.com/ContemplateLtd/filesystem-wrapper

My motivation was to be able to support long filenames (> 240 chars) on 
Windows, but this entails using the wide versions of the filesystem 
functions.

I based it on the patch that Alain posted a link to, but only supported 
the operations that we needed (openfile, opendir, readdir, closedir and 
is_directory). I also had to use an abstract type for pathnames to be 
able to handle the bizarre way that windows does long file names (you 
have to prefix the absolute name with '\\?\', as far as I can tell).

There is a little Makefile that assumes you are cross-compiling from 
Linux with the Debian-packaged cross compiler.

I am completely inexpert in Windows programming, so there are almost 
certainly bugs in it. It has been reasonably well tested with long 
filenames (we were doing static analysis of Java .class files, some of 
which are auto generated from XML Schemas and can have very long names), 
but I haven't tested it much on non-ASCII names. It converts back and 
forth between UTF-16 for Windows to UTF-8 for OCaml.

As Alain says, the full solution would be to fix OCaml itself.

Bob

On 05/02/16 10:54, Matthieu Dubuget wrote:
> Hello,
>
> I'm currently analysing a NTFS file-tree with a windows OCaml native application.
>
> This application is using:
> - Unix.{opendir,readdir,closedir}
> - and Unix.LargeFile.lstat
>
> The unix library of OCaml distribution is using ANSI variants of system functions. This is working fine until files or directories whose UTF-16 encoded name cannot be converted into the code page in use are reached.
>
> I'm about to write a small library to solve this problem: it would mimic the corresponding code from OCaml unix library, but using WIDE variants of microsoft system functions in the C stub instead of ANSI variants.
>
> Before going on: do you know of any library that already do this I could use?
>
> Thanks for any link.
>
> Salutations
>