Gauss-Legendre integration error

[davidbenjamin]

We use Gauss-Legendre integration in the strand bias model. The number of subdivisions increases with the read count and for very deep coverage this can cause a stack overflow because, unfortunately, Apache Commons has a very questionable recursive implementation.

The short-term fix is to cap the number of subdivisions. The long-term fix is to write some sort of simple adaptive 1D and 2D quadrature method. This ticket is for the short-term fix.

[mbabadi]

See my comment on issue #3318

[vdauwera]

Got another user reporting this problem -- this seems to be especially relevant for amplicon datasets where deep coverage is widespread.

[davidbenjamin]

@vdauwera This is a big deal. I'll put in a quick fix tomorrow evening.

[vdauwera]

Awesome thank you @davidbenjamin

[davidbenjamin]

Closed by #3335.