Arasan 21.1

[Dann Corbit]

I have found that there are lots of chess engines that do not work properly under Windows with UCI or Xboard.
It is almost always due to buffering.

I add about 4 lines of code to stop buffering and then all is well.
The Windows default (full buffering) is different than the Linux default (line buffering) so people who write their engines on Linux almost never put in code to do that.

Stockfish and Cfish are examples of programs that need this change (and many others as well, I just mention them because I have to change the code so often). Now, most of the time it works fine as is. But sometimes it will get decoupled. When that happens you will see crazy things in {for instance} Arena output, such as analysis output getting associated with the wrong position.

For instance, here is how I start off main() in Cfish:

Code: Select all

int main(int argc, char **argv)
{

    if( setvbuf( stdin, NULL, _IONBF, 0 ) != 0 )
        printf( "Incorrect type or size of buffer for stdin\n" );

    if( setvbuf( stdout, NULL, _IONBF, 0 ) != 0 )
        printf( "Incorrect type or size of buffer for stdout\n" );

    if( setvbuf( stderr, NULL, _IONBF, 0 ) != 0 )
        printf( "Incorrect type or size of buffer for stderr\n" );

    print_engine_info(0);

    int i = _flushall();

Line buffering would work as well as no buffering, but I do long time analysis so that there is no harm in using no buffering in my case.

[MikeB]

Thank you Jon. It's good to see you never run out of ideas.

May I ask you what value you recommend for syzygy probe depth if I use the full 6-man set and have them on SSD?

On an SSD drive , a setting of "1" should be fine. Assuming the higher setting delays the probing which is usually the case. If you notice too much of a decrease in nps, increase the setting.

[Zerbinati]

I have found that there are lots of chess engines that do not work properly under Windows with UCI or Xboard.
It is almost always due to buffering.

I add about 4 lines of code to stop buffering and then all is well.
The Windows default (full buffering) is different than the Linux default (line buffering) so people who write their engines on Linux almost never put in code to do that.

Stockfish and Cfish are examples of programs that need this change (and many others as well, I just mention them because I have to change the code so often). Now, most of the time it works fine as is. But sometimes it will get decoupled. When that happens you will see crazy things in {for instance} Arena output, such as analysis output getting associated with the wrong position.
Line buffering would work as well as no buffering, but I do long time analysis so that there is no harm in using no buffering in my case.

Very helpful Dan thank you!
I will implement the code in SugaR.

[Gabor Szots]

Thank you Jon. It's good to see you never run out of ideas.

May I ask you what value you recommend for syzygy probe depth if I use the full 6-man set and have them on SSD?

On an SSD drive , a setting of "1" should be fine. Assuming the higher setting delays the probing which is usually the case. If you notice too much of a decrease in nps, increase the setting.

Thanks Mike.

[jdart]

I have found that there are lots of chess engines that do not work properly under Windows with UCI or Xboard.
It is almost always due to buffering.

Is this an issue with Arasan? Arasan is C++ so it does this on startup:

Code: Select all

    setbuf(stdin,NULL);
    setbuf(stdout, NULL);
    std::cout.rdbuf()->pubsetbuf(NULL, 0);
    std::cin.rdbuf()->pubsetbuf(NULL, 0);

This hasn't changed for many versions.

--Jon

[pferd]

I have a small problem when I use arasan manually with the Linux command line. I have to type 'quit' twice in order to close it.

[Sven]

I have found that there are lots of chess engines that do not work properly under Windows with UCI or Xboard.
It is almost always due to buffering.

Is this an issue with Arasan? Arasan is C++ so it does this on startup:

Code: Select all

    setbuf(stdin,NULL);
    setbuf(stdout, NULL);
    std::cout.rdbuf()->pubsetbuf(NULL, 0);
    std::cin.rdbuf()->pubsetbuf(NULL, 0);

This hasn't changed for many versions.

--Jon

When using the second two calls (...->pubsetbuf()), the first two (setbuf(...)) are redundant IMO.

Setting standard input to unbuffered is crucial for chess engines. For standard output I restrict that to everything that is relevant for basic communication to the GUI while normal output during search (like PVs, info lines etc.) is excluded. The "basic communication" kind of output therefore is always followed by a "flush" call. In my engine this helps to reduce the penalty of frequent unbuffered search output in ultra-fast games.

[Dann Corbit]

I have found that there are lots of chess engines that do not work properly under Windows with UCI or Xboard.
It is almost always due to buffering.

Is this an issue with Arasan? Arasan is C++ so it does this on startup:

Code: Select all

    setbuf(stdin,NULL);
    setbuf(stdout, NULL);
    std::cout.rdbuf()->pubsetbuf(NULL, 0);
    std::cin.rdbuf()->pubsetbuf(NULL, 0);

This hasn't changed for many versions.

--Jon

No, Arasan behaves correctly.

[Dann Corbit]

I have found that there are lots of chess engines that do not work properly under Windows with UCI or Xboard.
It is almost always due to buffering.

I add about 4 lines of code to stop buffering and then all is well.
The Windows default (full buffering) is different than the Linux default (line buffering) so people who write their engines on Linux almost never put in code to do that.

Stockfish and Cfish are examples of programs that need this change (and many others as well, I just mention them because I have to change the code so often). Now, most of the time it works fine as is. But sometimes it will get decoupled. When that happens you will see crazy things in {for instance} Arena output, such as analysis output getting associated with the wrong position.
Line buffering would work as well as no buffering, but I do long time analysis so that there is no harm in using no buffering in my case.

Very helpful Dan thank you!
I will implement the code in SugaR.

For C++ programs, I also fiddle with the C++ I/O operations like this:

Code: Select all

int main(int argc, char* argv[]) {
    setvbuf( stdin, NULL, _IONBF, 0 );
    setvbuf( stdout, NULL, _IONBF, 0 );
    std::cout.setf( std::ios::unitbuf ); // For C++
    std::cin.setf( std::ios::unitbuf ); // For C++
    std::cout << engine_info() << std::endl;

[Zerbinati]

For C++ programs, I also fiddle with the C++ I/O operations like this:

Code: Select all

int main(int argc, char* argv[]) {
    setvbuf( stdin, NULL, _IONBF, 0 );
    setvbuf( stdout, NULL, _IONBF, 0 );
    std::cout.setf( std::ios::unitbuf ); // For C++
    std::cin.setf( std::ios::unitbuf ); // For C++
    std::cout << engine_info() << std::endl;

Thx Dann!