`Random` related problem?

[ThomasBreuer]

In the discussion of my issue gap-packages/walrus/issues/38 for the walrus package, @james-d-mitchell asked whether the problem is in fact in the Digraphs package. Meanwhile I think that the answer is yes.

The problem shows up in a test example that constructs RandomDigraph(100, 0.05). The underlying function is RANDOM_DIGRAPH, which calls the function rand from the standard C library. As far as I understand, its results are not reproducible in the sense that the random number generator is set to some default value in the beginning.
Note that GAP's Test function initializes the global random sources (GlobalMersenneTwister) in the beginning of the tests, in order to make the test results reproducible.

Is this behaviour of RANDOM_DIGRAPH intended?
Would it perhaps make sense to admit a random source as an optional argument of RandomDigraph, such that one can make test results reproducible?
In any case, the documentation of RandomDigraph should state what the meaning of "random" is.

[james-d-mitchell]

Thanks for the report @ThomasBreuer

Is this behaviour of RANDOM_DIGRAPH intended?

Probably this was just not thought through properly when it was implemented.

Would it perhaps make sense to admit a random source as an optional argument of RandomDigraph, such that one can make test results reproducible?

Yes, absolutely, that'd be great.

In any case, the documentation of RandomDigraph should state what the meaning of "random" is.

Agreed. We'd welcome PR's implementing these changes!

[ThomasBreuer]

So the current behaviour of RandomDigraph is regarded as intended (and should be kept), and the only problem is that the documentation should describe that RandomDigraph may yield not reproducible results?

If yes then I think the documentation could say that exactly those methods are affected that are based on the C function RANDOM_DIGRAPH; in the other methods, the GAP function Random is called, which uses GlobalMersenneTwister if no random source is given as an argument.

(Thus is is easy to support an optional random source for the RandomDigraphs calls that are not based on RANDOM_DIGRAPH, but I do not know what the best solution would be for the remaining cases.)

Alternatively, one could consider getting rid of the rand calls, but this would be a breaking change, since the non-reproducibility of results would not be reproducible any more.