On 15.06.2015 16:52, William A Rowe Jr wrote:
Great report, thanks Bert. I'll review.
I verified the patch with trunk on Windows 10/VS2015; looks good.
On Mon, Jun 15, 2015 at 7:19 AM, Bert Huijben <firstname.lastname@example.org> wrote:
Trying yet another e-mail address... Nice that this just bounces instead of being moderated :(
> -----Original Message-----
> From: Bert Huijben [mailto:email@example.com]
> Sent: maandag 15 juni 2015 14:04
> To: firstname.lastname@example.org
> Cc: Stephen White; email@example.com
> Subject: [Patch] apr_file_path_merge returns incorrect truepath on certain
> 'short paths' on Windows
> In the last few months we got two reports from Subversion users that
> passing certain short paths to 'svn' failed for them.
> > mkdir ALongName.impl
> > dir /x
> 27/03/2015 12:50 <DIR> ALONGN~1.IMP ALongName.impl
> And then passing this 'ALONGN~1.IMP' path works just fine.
> > mkdir api.impl
> > dir /x
> 27/03/2015 12:52 <DIR> API~1.IMP api.impl
> And then passing the 'API~1.IMP' path doesn't (uses api.implP)
> (Samples copied from a mail from Stephen White to firstname.lastname@example.org)
> The problem here is that the short path is actually longer than the long
> path. which triggers a bug in APR's truename code that we use. In this
> specific case the length of the updated path is not properly updated, to
> reflect the shorter total path length.
> When trying to write a testcase for this issue I found that Windows
> disables short path generation on user created NTFS filesystems these
> days. Only on system partition the default setup enables short path
> creation. (An admin can enable short paths via something like 'fsutil
> 8dot3name set R: 1', or a flag can be passed to format.exe)
> The attached patch fixes the problem, which appears to be specific for the
> last path component (saveslash = FALSE).
> I would like to see this problem fixed in apr 1.5.
> The attached patch applies directly to 1.5.x and trunk.