Author: Eugene Nalimov
Date: 21:17:24 12/01/01
Go up one level in this thread
Welcome to the real world, Cristophe! Surprise -- there is no stupid thing the user will not do. And of course user will do the things no developer or tester will never think about. I can assure you that at least half of the MS software problems happen because of that :-) Problem with incomplete TBs sets is well-known, and solution is also well known. If you decided that you don't want to implement those properly, please live with the consequences of your decision. Of course you documented that problem in the documentation, right? If so you can ask testers to remove the results. Now regarding your question: when I'll again write the chess program, I will do 2 things regarding hash table size: (1) Set the minimal hash size to something reasonable, e.g. 4Mb [I am talking about version of the program running on the desktop Windows computer]. (2) If user allocates for the hash tables less than e.g. 1/4 of the physical memory on the computer for the long time control game, give him a warning. I believe one of your favorite complains about Windows is that you are uncomfortable with default settings. Now you can look at exactly that problem from the other side -- somebody [you] can be very unhappy with the default settings of *your* program. Also you can find out that giving too much control to users can result in the high support costs and to the lost productivity of the *users*. Eugene On December 01, 2001 at 23:06:28, Christophe Theron wrote: >On December 01, 2001 at 18:06:49, Uri Blass wrote: > >>On December 01, 2001 at 17:27:31, Harald Faber wrote: >> >>>On December 01, 2001 at 10:13:54, Bertil Eklund wrote: >>> >>>>On December 01, 2001 at 09:52:54, Harald Faber wrote: >>>> >>>>>On November 30, 2001 at 18:59:45, Bertil Eklund wrote: >>>>> >>>>>I can implement the code for mating with b+n or I can let it out and say "the >>>>>TBs solve this". Christophe obviously did the 2nd, who is to blame? >>>> >>>>I think that most users play with the delivered 3 and 4 man tablebases. There >>>>are also a lot of users that plays with the four Turbo-cds from Chessbase or >>>>Millenium, in fact it is probably a houndred times more people that only use the >>>>standard tbs that is included with the programs. >>>> >>>>Bertil >>> >>>Sorry to say that but it should be common knowledge that using incomplete >>>5-man-TBs produces such behaviour and therefore unrealistic results. >> >>It should be common knowledge of programmers to use the incomplete 5 piece >>tablebases correctly(stopping to use tablebases when the distance to mate is not >>reduced is a possible way). >> >>Uri > > > >I'm sorry but this is the responsability of the testers. > >It is common knowledge of programmers that 12Kb of hash tables is inferior to >192Mb. > >So if the tester wants to use 12Kb of hash table, should the program: >1) use 12Kb? >2) refuse to run? >3) use 192Mb no matter what? > >I can already hear the complains if I program 2 or 3... > >There are many ways to set up a program incorrectly. The programs generally are >able to catch a number of common cases, but not all of them. > >And some of the incorrect setups for long time controls can be absolutely >correct in some other circumstances (tests on positions only, bullet games...). > >What is the SSDF testing? The real strength of chess programs or their ability >to recover from bad setups? > > > > Christophe
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.