Polynomial selection very strange in SVN 1030M

[fivemack]

I'm trying to run some polynomial selection on a C194 with an updated msieve, and it is choosing a much sparser set of initial coefficients (some ranges of 10000 c5 have no valid initial coefficient) and then wanting to run each coefficient for a hundred days! Maybe 1600 seconds was a bit short, but 8640000 seconds is implausibly long.

[Gimarel]

I guess this is because of a patch to yafu/msieve that I posted. I use this to factor smaller composites and did not care for composites above C160.

I've set the timeout so big to disable it because in my experience it's better to set the stage1_norm to a value that stage1 can finish in time.

The high_coeff_multiplier is set to 360 because in my experience the overall chance to find a good poly is better with a smoother leading coefficiant. But most likely this is not the optimal value.

[xilman]

Originally Posted by Gimarel

I guess this is because of a patch to yafu/msieve that I posted. I use this to factor smaller composites and did not care for composites above C160.

Wah!

Some of us very much care.

[Gimarel]

Originally Posted by xilman

Wah!

Some of us very much care.

I do polyselect for larger numbers but I don't have generic parameters.

The patch was never intended to be incorporated into msieve, at least not for parameters above C160.

[jasonp]

The configuration for large polyselect jobs does many fairly outlandish things; I guess it's worth revisiting how configuration works so that we can get sensible defaults even without user-specific configuration.