Serious ECMNet Client/Server 2.5.7 bug

[rogue]

I should kick myself for releasing code with this bug.

In the 2.5.7 and 2.5.6 releases there is a bug in Work.cpp that will cause the client to rarely (if ever) find a factor. If you running with the new client, you will have to fix this problem immediately.

Until I release 2.5.8, this is what you have to do. In Work.cpp in the procedure SetUpCall, look for the line:

sprintf(s_Command, "%s -sigma 2937239194 %s < %s > %s", s_GMPExe, temp1, s_InWorkFileName, s_OutWorkFileName);

change it to

sprintf(s_Command, "%s %s < %s > %s", s_GMPExe, temp1, s_InWorkFileName, s_OutWorkFileName);


I am seriously embarassed by this bug and am very sorry for the ECM work that will be lost as a result. I myself lost over 1 year of ECM factoring because of this. I will let you know when 2.5.8 is released.

--Mark Rodenkirch

[frmky]

Version 2.5.6 is ok. I've been running it for quite a while now on the XYYXF composites, and have found quite a number of factors. In fact, it's running right now with the following work_a.out:

GMP-ECM 6.0 [powered by GMP 4.1.4] [ECM]
Input number is 32...19 (219 digits)
Using B1=1000000, B2=839549779, polynomial Dickson(3), sigma=3798466865

so sigma isn't at 2937239194 in that version.

Greg

[rogue]

Thanks. It's nice to know that only that release has the problem. That limits the scope significantly.

[frmky]

It looks like I may have spoken too soon! Were there two different 2.5.6 releases? The one I compiled right after the initial release does not have the bug, but the one I downloaded on Feb 17 and compiled on my Linux machine does have the bug! Both claim to be 2.5.6. Hmmm...

[rogue]

Originally Posted by frmky

It looks like I may have spoken too soon! Were there two different 2.5.6 releases? The one I compiled right after the initial release does not have the bug, but the one I downloaded on Feb 17 and compiled on my Linux machine does have the bug! Both claim to be 2.5.6. Hmmm...

You probably had a pre-release which I modified before submitting to Paul. Paul's site still says that both are probably affected by the bug. I hope that anyone who is using the new client has d/l'd 2.5.8.

[antiroach]

ive got a quick question about how the behavior of the ecmnet server (2.5.8). When a person is assigned a number are they the only ones who can be working on it or is it possible for multiple people to be assigned the same number to work on at the same time?

[rogue]

Originally Posted by antiroach

ive got a quick question about how the behavior of the ecmnet server (2.5.8). When a person is assigned a number are they the only ones who can be working on it or is it possible for multiple people to be assigned the same number to work on at the same time?

Yes, multiple people can have the same assignment. Use the strategies in the ecmserver.cfg file to reduce the likelihood of that happening.