strcspn(3) complexity improvement

[Brooks Davis]

On Wed, Mar 30, 2005 at 09:06:13PM +1000, Peter Jeremy wrote:
> On Wed, 2005-Mar-30 10:34:35 +0200, Jeremie Le Hen wrote:
> >Andreas Hauser made a patch to strcspn(3) for the DragonFly project
> >which makes it faster when dealing with long strings [1] (rev 1.4).
> >It basically changes the complexity of the function from
> >    O(strlen(str) * strlen(chars))
> >to
> >    O(strlen(str) + strlen(chars))
> >by using a charset.
> 
> It has a significantly higher overhead due to the need to zero the charset.
> 
> >I have two questions.  First, is this change worth enough to be merged
> >in FreeBSD (this function is currently used in 42 binaries from
> >/{,usr/}{s,}bin) ?  I mean does the performance gain on large strings
> >compensates the use of a large 256-bytes buffer ?
> 
> You are proposing this change so I think it's up to you to demonstrate
> an improvement.  I don't think the space is an issue in userland (it
> would be in the kernel) so it's just a matter of which is faster.  Did
> Joerg or Andreas provide any performance test results?  My gut feeling
> is that strcspn() isn't heavily used enough or with long enough
> "chars" arguments for the change to be noticable in the base system.

The real question I have is, how long does the string need to be before
this is a win and how much does it hurt for typical string lengths?
I've written code with strcspn that needed to perform well, but it was
parsing 80-column punch card derived formats.

-- Brooks

-- 
Any statement of the form "X is the one, true Y" is FALSE.
PGP fingerprint 655D 519C 26A7 82E7 2529  9BF0 5D8E 8BE9 F238 1AD4
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-current/attachments/20050330/e46dedf4/attachment.bin