[Pkg-octave-devel] Fwd: Bug#876777: octave-stk FTBFS on i386: test failure

[Julien Bect]

Hi DOG,

octave-stk has been marked for autoremoval from testing because of the 
following test failure on unstable/i386:

Source: octave-stk
Version: 2.4.2-1
Severity: serious

https://tests.reproducible-builds.org/debian/history/octave-stk.html
https://tests.reproducible-builds.org/debian/rb-pkg/unstable/i386/octave-stk.html

...
***** test
  model = stk_model ('stk_materncov_iso');
  [param0, model.lognoisevariance] = stk_param_init (model, xi, zi, BOX);
  model.param = stk_param_estim (model, xi, zi, param0);
  zp = stk_predict (model, xi, zi, xt);
  assert (max ((zp.mean - zt) .^ 2) < 1e-3)
!!!!! test failed
operator *: nonconformant arguments (op1 is 2x2, op2 is 3x1)
shared variables
   scalar structure containing the fields:
...
Summary: 1317 tests, 1316 passed, 0 known failures, 0 skipped
/usr/share/cdbs/1/class/octave-pkg.mk:108: recipe for target 'check-pkg' failed
make: *** [check-pkg] Error 1


I have been able to reproduce the problem, using cowbuilder, for the 
i386 architecture (the problem doesn't seem to exist on amd64).

This problem is not directly in octave-stk.  In fact, the problem 
happens when sqp calls qp with a matrix that is not sufficiently 
positive definite.

Both functions (sqp, qp) are in Octave itself, and the problem is 
already known : http://savannah.gnu.org/bugs/?38875.

How should we deal with this issue to prevent octave-stk-2.4.2-1 from 
being removed from testing ?

@++
Julien

PS: this test is in ./paramestim/stk_param_init.m