Re: Testing program with random output [newbie]

[Robert Hartmann]

Hi Ben and others,

Am 25.07.2013 00:07, schrieb Ben E.:
> Hi Robert
>
> > Is it possible to create a test runnable with dejagnu for checking a
> > commandline-program which has randomized output: Testing concrete
> > single threaded commandline java programs which use Math.random().
>
> Sure, why not?  If you can express verbally how you decide that the
> test has passed, it can be implemented in Tcl.  You can use something
> like "expr rand()" to generate random numbers.
>
> Ben

First the program, which should be checked via DejaGNU, produces
an arbitary number of strings (ending with newline).

Well ok, I could verify by "hand" if the output (the randomized 
generated string) is valid.

For validating one string (= one line) I need two integer variables, say 
A and B.

The output string only containts the symbols  < > ^ v -
 and four defined words, say "WordA" "WordB" "WordC" "WordD".

Now I have to parse the string, and modify A=0 and B=0 this way:

For every
   < : A--,
   > : A++,
   ^ : B++,
   v : B--,
   - : do nothing

A string called "valid" if:
  B<0 and "WordD" is the last substring,
  B>0 + -2<A<2 + "WordC" is the last substring,
  A>=2 + "WordB" is the last substring,
  A<=2 + "WordA" is the last substring.
All other strings are "invalid"


If I look at my dejagnu "echo word", then there are lines like:

========================
[...]
proc PROGRAM_test {expr result} {
        global feedback
        send "$expr\n"
        expect {
          -gl "$result\r\n" {pass "$expr"}
          "*\r\n" { fail "$expr"}
          timeout { fail "timeout $expr"}
        }
}

PROGRAM_start
PROGRAM_test "Test1" "Echo: Test1"
PROGRAM_test "Test2" "Echo: Test2"
send_user "$feedback\n"
PROGRAM_exit
========================


For shure I have to modify the procedure PROGRAM_test for
validating PROGRAM output. But I don't get it.

I did not see, how I can do my string parsing in the "expect"-language,
and I don't see, how I can check for an indeterministic number of lines.


Best regards,
Robert