Date: Sun, 30 May 2010 19:14:29 -0400
From: Christopher Faylor <cgf-use-the-mailinglist-please AT cygwin DOT com>
To: cygwin AT cygwin DOT com
Subject: Re: Cygwin Performance and stat()
Message-ID: <20100530231429.GB21673@ednor.casa.cgf.cx>
Reply-To: cygwin AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
References: <efe8a37b2e4466daa7b6eb1aa610c3d7 DOT squirrel AT www DOT webmail DOT wingert DOT org> <20100530170747 DOT GA8605 AT ednor DOT casa DOT cgf DOT cx> <f460895a8fc53da26cb91259a4005da2 DOT squirrel AT www DOT webmail DOT wingert DOT org> <20100530210321 DOT GB11159 AT ednor DOT casa DOT cgf DOT cx> <4C02DE2D DOT 90600 AT t-online DOT de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <4C02DE2D.90600@t-online.de>
User-Agent: Mutt/1.5.20 (2009-06-14)
Mailing-List: contact cygwin-help AT cygwin DOT com; run by ezmlm
Precedence: bulk
Sender: cygwin-owner AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com

On Sun, May 30, 2010 at 11:52:45PM +0200, Christian Franke wrote:
>Christopher Faylor wrote:
>> On Sun, May 30, 2010 at 12:51:31PM -0700, Christopher Wingert wrote:
>>    
>>> I assume POSIX compatibility.  However, I bet there are cases where one
>>> can sacrifice compatibility for performance (configurable with an
>>> environment flag of course).
>>>
>>>      
>
>The problem is that POSIX stat() requires to provide all information in 
>struct stat. This is expensive: convert ACL into mode, convert filetimes 
>into time_t, lookup uids, invent ino, ....
>Typically applications don't need all info but there is no way to tell 
>this to stat().
>
>>> See
>>>
>>> http://marc.info/?l=git&m=122278284210941
>>>
>>> for an example.
>>>      
>> Yes, I got what you meant.  I think that bypassing Cygwin, assuming you
>> can do something good enough is a remarkably bad idea.
>
>A probably better idea would be to add another stat() variant (this is 
>IMO missing in POSIX) to Cygwin which allows to specify which info is 
>needed.
>
>Something like:
>
>enum {
>   CYGSTAT_MODE = 0x01,
>   CYGSTAT_INO = 0x02,
>   CYGSTAT_SIZE = 0x04,
>   CYGSTAT_ATIME = 0x08,
>   ...
>};
>
>int cygwin_stat4(const char *path, struct stat *buf, unsigned needed, 
>unsigned *pvalid);
>
>Where 'needed' contains all CYGSTAT_* flags for the stat fields needed. 
>If 'pvalid' is != 0, the flags for the valid fields are returned in the 
>variable. This can be a superset of 'needed' if some info is available 
>for free.
>
>Usage example:
>
>// Get st_size only.
>#ifdef __CYGWIN__
>#define stat_size(p, b) cygwin_stat4(p, b, CYGSTAT_SIZE, NULL)
>#else
>#define stat_size(p, b) stat(p, b)
>#endif

And then you'd have to demonstrate that such an implementation actually
showed a measurable improvement, that it doesn't impact the existing
stat(), and that it would actually be used.

And, of course, as soon as you start talking about implementing
something non-POSIX you have strayed away from the whole point of
Cygwin.

And, also of course, we can get one new idea per message here in the
cygwin mailing list but that doesn't actually cause anything to be
implemented.

cgf

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