Uh, yet another result.

@Avinoam: strange, I see

octave:3> test_sqp
x =

  -1.71714
   1.59571
   1.82725
  -0.76364
  -0.76364

obj =  0.053950
info =  101
iter =  9
nf =  9
lambda =

  -0.0401627
   0.0379578
  -0.0052227

@Doug: does my patch work for you?

This discrepancy was introduced in #32008, comment #4. If you comment the break just after info = 104;, you see that at the next iteration sqp terminates with 101. I think it is a bit too strict to exit with 104 as soon as delx < tol * x, whitout testing whether x_new and company satisfies the problem. What about the attached proposal?

(file #44815)

Running the examaple given in sqp documentation (test_sqp.m, attached)

I get:

>> test_sqp
x =

  -1.71714
   1.59571
   1.82725
  -0.76364
  -0.76364

obj =  0.053950
info =  104
iter =  8
nf =  9
lambda =

  -0.0401627
   0.0379578
  -0.0052227

info = 104 (The stepsize has become too smal), nf = 9,
while in the documentation info=101 (The algorithm terminated normally), and nf=10.

Is it a documentation error or function error?