Re: New stat stuff (was [PATCH] improve performance of stat() ope rations (e.g. ls -lR ))

[Christopher Faylor <cgf at redhat dot com>]

On Wed, May 29, 2002 at 08:30:09PM -0400, Don Bowman wrote:
>
>> On Tue, May 28, 2002 at 10:25:37PM -0400, Christopher Faylor wrote:
>> >On Tue, May 28, 2002 at 09:58:52PM -0400, Don Bowman wrote:
>> >>
>> >>So I've performed a mini-benchmark of Chris' changes.
>> >>
>> >>I did a ls -lR >/dev/null of the cygwin source tree on my
>> >>notebook.
>> >>
>> >>Baseline (current setup.exe install): 1m14.9s
>> >>'statquery' patch I sent earlier: 4.081s
>> >>Current CVS tree: 3.718s
>> >>Current CVS tree w/ -E switch to mount: 3.711s
>> >>Current CVS tree w/ -X switch to mount: 3.716s
>> >>
>> >>Not all that scientific, I ran each twice, took the 2nd timing.
>> >>So, looks good, excellent work. I still don't see any 
>> >>difference on the -E or the -X tho'.
>> >
>> >That has got to mean that there's something wrong in the stat
>> >logic.  I didn't do anything to speed up the normal case, AFAIK,
>> >unless you're doing this on a FAT/FAT32 partition.
>> 
>> Actually, even in that case, it shouldn't make that big a deal.
>
>Its the anti-virus. Your change no longer opens the file for read,
>so the anti-virus doesn't do anything, thus the enormous difference.
>As to why the negligible difference between -E/-X/nothing, no
>idea.

No, my change should not have stopped the file from being opened
in the normal case.  It was only in the -E, -X cases that it should
have an effect.

I did have a logic error which is corrected in the latest snapshot.

cgf