[Help-glpk] CPLEX vs GMPL: different answers for same problem

[Reginald Beardsley]

I've successfully converted from CPLEX to GMPL but have discovered that the 
answers are different.  They're both excellent, but they're not the same.

My problem is quite floating point sensitive, so I'm using --exact for both.

I *think* I'm solving the same problems, but the input to glpsol is being 
created by a program and the files run 50,000+ lines so it's quite difficult to 
check. That was actually the main motivation for switching. The programs are 
the same except for the output format. 

Does anyone know of a reason they would be different?  The solution errors are 
very small in both cases, but I'd like to understand why they're not the same.

Thanks,
Reg