Message-ID: <46B0D18D.40409@tlinx.org>
Date: Wed, 01 Aug 2007 11:31:41 -0700
From: Linda Walsh <cygwin AT tlinx DOT org>
User-Agent: Thunderbird 1.5.0.12 (Windows/20070509)
MIME-Version: 1.0
To: cygwin AT cygwin DOT com
Subject: Re: unlink()'s not quite POSIX behavior.
References: <f8o1qr$jtk$1 AT sea DOT gmane DOT org>
In-Reply-To: <f8o1qr$jtk$1@sea.gmane.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Mailing-List: contact cygwin-help AT cygwin DOT com; run by ezmlm
Sender: cygwin-owner AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com

Joe Smith wrote:
>> When the file's link count becomes 0 and no process has the file open,
>> the space occupied by the file shall be freed and the file shall no 
>> longer
> Could we at least simulate the behavior by moving the file out of the 
> way (simultaionsly renaming it to something unique),
> and forcing it into the delqueue? (by Setting the 
> FILE_DISPOSITION_INFORMATION's DeleteFile flag to true?)
> 
> Wait a moment, that looks to be exactly what unlink_nt is doing?
> 
> (The problem is with a call in Python3k getting a "Permission denied" 
> (ERRNO 13) error when attempting to create a file shortly after it has 
> been deleted with unlink. That seems to be consistant with standard 
> windows behavior for deleting a file, as trying to create it again 
> before the last handle is closed would return an ERROR_ACCESS_DENIED.)
> 
> However, it looks like with unlink_nt that should not be happening, right?
 >
> Nevermind. I see that the relavent changes to NT_unlink are just very 
> recent, and the only problem is that the latest cygwin dll does not 
> incldue those changes yet.> 
-----

	I seem to remember being told that changing the rename function
to work in this, POSIX-compatible way, was impossible with cygwin's current
fork implementation -- and that Cygwin would have to keep track of
what files have been renamed (if any were DLL's) and propagate changes
to child processes (after a fork call) so they could correctly pull
in contents of "old, replaced DLL's".

	Is this change to fork something that is being worked on?  I.e.
will we be able (at an application (setup.exe or cygperl replacing DLL's)
level) to rename "in-use" DLL's and copy in new versions?

	I thought Eric indicated that this tracking would make an already
slow process even slower(?)  and wasn't worth it.
(http://sourceware.org/ml/cygwin/2006-12/msg00906.html)

Has this changed?  Or, more accurately, is this about to change? Was
it decided to go with the (seemingly) POSIX correct action to allow
the "rename-to-tmp+copy-in-new-DLL" (Rtt+CinD) to work?  Will
Cygwin be keeping track of the old "deleted" (but still in-use
via a cygwin open Filehandle) libraries to duplicate into new
children processes?

	I seemed to remember feeling my "hand slapped" at suggesting
we use "Rtt+CinD" as it wouldn't be practical to implement.
Maybe the open-handle->"deleted file" idea isn't that expensive
to implement afterall?  (Despite it being a RPITA; hail emperor bill).


Linda

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/