Fastest String Search Algorithm.

[Adhemerval Zanella]

On 09/06/2021 09:22, Amit Choudhary wrote:
> On Wed, Jun 9, 2021, 5:41 PM Adhemerval Zanella <
> adhemerval.zanella@linaro.org>
> 
>>
>> Indeed, as Siddhesh has noted I agree that a better bench-strstr with more
>> meaningful tests is the start point to evaluate a newer strstr
>> implementation.
>>
> 
> I invented my algorithm because I thought about it.
> 
> Getting my algorithm in glibc is not my goal.
> 
> I can defintely provide many real world haystack and needle in a file.
> 
> But, currently, I don't have much time to change strstr() benchmarks.

The problem is really hard to judge any improvement if we can't readily
verify and replicate the results in a consistent manner.  The benchtests
we have is not perfect, but it is a starting point so any developer and
reviewer can verify and play with the proposed change.

You outlined possible limitations of current benchtests, so I would
ask you if could improve the tests so it would make the review and
test easier.

Throwing numbers and external projects (specially ones that does not
seem to have much sense such as java source code) it is not a way to
discuss a potential performance improvements. Usually reviewers, including
myself, won't spend any time trying to adjust glibc code to check your
claims.

Also, I would like to ask to please work and listen to Wilco suggestions.
He was the one that code the current strstr implementation and he did a
bit fair or research on the topic before come with a newer implementation.