src/string/strstr.c, src/string/wcsstr.c: wrong estimates for MIN(l,63)?

src/string/strstr.c, src/string/wcsstr.c: wrong estimates for MIN(l,63)?

[Yves Bastide]

Hi,

twoway_strstr and twoway_wcsstr seem to contain a bogus estimation:

/* Fast estimate for MIN(l,63) */
size_t grow = l | 63;

"grow" is thus always at least 63...

Regards,

Yves

Re: src/string/strstr.c, src/string/wcsstr.c: wrong estimates for MIN(l,63)?

[Rich Felker]

On Fri, Mar 21, 2014 at 03:07:22PM +0100, Yves Bastide wrote:
> Hi,
> 
> twoway_strstr and twoway_wcsstr seem to contain a bogus estimation:
> 
> /* Fast estimate for MIN(l,63) */
> size_t grow = l | 63;
> 
> "grow" is thus always at least 63...

If you look at how grow is used on the very next line, this should
make sense.

Rich

Re: src/string/strstr.c, src/string/wcsstr.c: wrong estimates for MIN(l,63)?

[Szabolcs Nagy]

* Rich Felker <dalias@aerifal.cx> [2014-03-21 12:48:17 -0400]:
> On Fri, Mar 21, 2014 at 03:07:22PM +0100, Yves Bastide wrote:
> > /* Fast estimate for MIN(l,63) */
> > size_t grow = l | 63;
> > 
> > "grow" is thus always at least 63...
> 
> If you look at how grow is used on the very next line, this should
> make sense.

the comment is wrong though
z grows by at least max(63,l) unless end-of-string is reached

Re: src/string/strstr.c, src/string/wcsstr.c: wrong estimates for MIN(l,63)?

[Rich Felker]

On Fri, Mar 21, 2014 at 06:07:45PM +0100, Szabolcs Nagy wrote:
> * Rich Felker <dalias@aerifal.cx> [2014-03-21 12:48:17 -0400]:
> > On Fri, Mar 21, 2014 at 03:07:22PM +0100, Yves Bastide wrote:
> > > /* Fast estimate for MIN(l,63) */
> > > size_t grow = l | 63;
> > > 
> > > "grow" is thus always at least 63...
> > 
> > If you look at how grow is used on the very next line, this should
> > make sense.
> 
> the comment is wrong though
> z grows by at least max(63,l) unless end-of-string is reached

Yes, (l|63) is >= both l and 63, so it should read MAX, not MIN. The
idea is that it's useless to grow by less than l, but it's inefficient
to grow by less than some fixed threshold even if the needle is tiny.

Rich

Re: src/string/strstr.c, src/string/wcsstr.c: wrong estimates for MIN(l,63)?

[Yves Bastide]

2014-03-21 18:44 GMT+01:00 Rich Felker <dalias@aerifal.cx>:
> On Fri, Mar 21, 2014 at 06:07:45PM +0100, Szabolcs Nagy wrote:
>> * Rich Felker <dalias@aerifal.cx> [2014-03-21 12:48:17 -0400]:
>> > On Fri, Mar 21, 2014 at 03:07:22PM +0100, Yves Bastide wrote:
>> > > /* Fast estimate for MIN(l,63) */
>> > > size_t grow = l | 63;
>> > >
>> > > "grow" is thus always at least 63...
>> >
>> > If you look at how grow is used on the very next line, this should
>> > make sense.
>>
>> the comment is wrong though
>> z grows by at least max(63,l) unless end-of-string is reached
>
> Yes, (l|63) is >= both l and 63, so it should read MAX, not MIN. The
> idea is that it's useless to grow by less than l, but it's inefficient
> to grow by less than some fixed threshold even if the needle is tiny.
>
> Rich

OK, understood.

Thanks!

yves