Re: Cygwin mentioned in "current directory" discussion

[Andy Koppe <andy.koppe@gmail.com>]

On 15 November 2010 19:26, Corinna Vinschen wrote:
> On Nov  9 23:37, Christopher Faylor wrote:
>> https://blogs.msdn.com/b/oldnewthing/archive/2010/11/09/10087919.aspx
>
> This morning I added a comment to the blog, #42 in the list:
>
> --- SNIP ---
>  [I wouldn't put it past the Cygwin guys to patch ntdll...-Joshua]
>
>  That's a bit over the top, but the new, undocumented way Vista and later
>  uses to set the CWD requires some serious hacking to get the Linux semantics
>  right.
>
>  Note that the problem is not that SetCurrentDirectory creates a
>  directory handle.  The problem is that it creates a directory handle
>  without the chance to set certain flags, analogous to the CreateFile
>  call:
>
>  - SetCurrentDirectory always opens dirs without the FILE_SHARE_DELETE flag.
>  - SetCurrentDirectory always opens dirs without the FILE_FLAG_POSIX_SEMANTICS flag.
>  - SetCurrentDirectory always opens dirs without the FILE_FLAG_BACKUP_SEMANTICS flag.
>
>  Apart from that, an annoying problem of the SetCurrentDirectory call is
>  it's quite obvious flaw to allow only paths of up to 258 chars, even the
>  UNICODE variant.  That's due to the fact that the CWD path storage is a
>  fixed buffer of 260 wide chars (including a trailing backslash and a
>  L'\0').  This is still true for the Vista/W7 style of CWD storage on the
>  heap, even though the method would easily allow longer paths.
>
>  A transparent solution for Windows 8 and later would be an Ex call as
>  proposed by Andy, like this:
>
>    BOOL SetCurrentDirectoryEx (LPCTSTR lpDirectoryName,
>                                DWORD dwShareMode,
>                                DWORD dwOpenFlags)
>
>  with SetCurrentDirectoryExW allowing the same long pathname syntax for
>  paths up to 32K as the CreateFileW and similar calls.
>
>  Actually, for Cygwin it would already be sufficient (and even more
>  convenient), to export the global pointer pointing to the current CWD
>  datastructure on the heap.  Or better, to export the function which
>  returns this pointer.  This would allow to get rid of the awkward code
>  which extracts this global pointer by scanning the
>  RtlGetCurrentDirectory_U function.
>
>  I guess this should go without saying, but it would be nice if this new
>  CWD storage method would be officially documented...
> --- SNAP ---
>
> Two minutes ago, I found that my comment had been silently removed!
> I'm not aware that my comment contained offensive, abusive, or
> disrespectful language.  I'm really wondering why Raymond decided
> to remove my comment.  This is quite disappointing.

Perhaps it got swallowed by a COM server that can't deal with a
disappeared working directory. ;)

Andy