Starting point for "SROSENBR" differs from the .SIF file.

[NicolasL-S]

This seems like a strange CUTEst bug, but I thought I would ask here just in case.

In the "ROSENBR" problem, the starting point is [-1.2, 1.0]. This is in agreement with the code in the file ROSENBR.SIF:

START POINT

ROSENBR X1 -1.2
ROSENBR X2 1.0

In the file SROSENBR.SIF, the starting point is defined with the following code

START POINT

DO I 1 N/2
IM 2I I 2
IA 2I-1 2I -1
X SROSENBR X(2I) 1.0
X SROSENBR X(2I-1) -1.2
ND

Hence, I would expect the starting point to be [-1.2, 1.0, -1.2, 1.0, -1.2, 1.0, -1.2, 1.0...]. But surprisingly, when I run the following code

using CUTEst
CUTEst_nlp = CUTEstModel("SROSENBR")
CUTEst_nlp.meta.x0

, I get the starting point

julia> CUTEst_nlp.meta.x0
5000-element Vector{Float64}:
 1.2
 1.0
 1.2
 1.0
 1.2
 1.0
 1.2
 1.0
...

[amontoison]

We call the routine csetup in the Julia interface so I don't think that the bug is in CUTEst.jl.
I need to check if the issue is in the SIF decoder or in the Fortran routine.
@nimgould Do you have an idea of the culprit?

[nimgould]

Yup, it is a bug in the SIF file, the -1.2 is shifted one space to the left, and SIF requires numbers to lie between very specific field limits. Since tis file has been around since 1989, I guess we will have to live with this starting point, but as SIF allows more than one start, I'll add the correct one as a second. Thanks for reporting this