Note the upstream numbl fix beside the slice-orientation workaround

magland <jeremy.magland@gmail.com> committed 1 month ago commit ee7257c73485 parent aa03146 Browse files

1 changed file+4−2

matlab/maraca_hankel.mmodified+4−2View file

@@ -22,8 +22,10 @@ for n = Lstart:-1:1

2222 jj = jj * 1e-250;

2323 end

2424 end

25-% The explicit second index works around a numbl slicing bug that can hand

26-% back jj(1:lmax+1) as a row vector depending on what else the script calls.

25+% The explicit second index pins the column orientation. numbl <= 0.4.19

26+% guessed a row here when the JIT could not prove jj was a column (its

27+% length is only known at runtime), silently transposing the result; fixed

28+% upstream, but kept until a release carrying the fix is out.

2729 j = jj(1:lmax + 1, 1) * ((sin(x) / x) / jj(1));

2830

2931 % ---- y_l: upward recursion ----