simplex algorithm - R. Agnew function

Can I persuade anyone to look through R. Agnew's implementation of the

simplex LP algorithm (APL quote-quad, March 1985; correction of some text in

comments, September 1985) and re-assure me that the completion status should

**not** depend on the value of APLs random link, []RL? (As far as I can see,

[]RL appears only in lines 50 and 52 of the published version of the function).

The reason for this odd request is that I've been trying a small problem using

the function, and I found that repeats of the same problem gave different

completion results -- most often 0 (optimal solution found) but about 25% of

the time it returned a value of 2, which means an unbounded solution. After

lots of experimentation I seem to come up with the result that []RL can affect

theresult. I hard-coded in a value which seems to work (1390765147) and was

able to run 150 repeats of the problem and get the same completion code each

time.

System used: Manugistics APL*Plus v.10.1. The problem appeared on

two different pcs, each with a 33 mhz '486DX.

At some point I'll try the same thing on APL-Plus/III and see what happens.

If anyone would like a workspace to experiment with, in APL*PLUS/PC (.aws)

form, let me know, and I'll send it by email.