Roughly the idea there is to re-use random elements for different purposes: if a random element was generated and used by naming, there is no reason why later on it should not be used again for constructive recognition. To handle this, recog maintains a stream of random elements, plus a bunch of "pointers" or "indices" into this stream, for different purposes. The challenge is to properly document this, and then use it.
Some more details are in this code comment:
# RandomElm and RandomElmOrd take a recog record, a string, and a
# bool as inputs.
# The string is used as a stamp to request a random element or order for a
# specific computation. RandomElm and RandomElmOrd will first try to reuse
# random elements and orders generated with different stamps.
# For example, if a computation which used stamp := "A" has already computed
# random elements or orders, then RandomElm and RandomElmOrd will reuse these
# if called with stamp := "B".
#
# The components of the recog record involved are explained in
# RecogNode.
#
# HACK: For recog records created by RecogNode the method
# RandomElm is by default stored in the component ri!.Grp!.pseudorandomfunc.
# A method for PseudoRandom is installed such that it calls
# RandomElm(ri, "PseudoRandom", false).
Once all this is done, we could perhaps also get rid of the PseudoRandom hack we have in place to get existing code to at least somewhat use these helpers:
InstallMethod( PseudoRandom, "for a group object with generators, use func",
[ IsGroup and HasGeneratorsOfGroup ], 1,
function( g )
local l;
# FIXME: get rid of this hackish override of PseudoRandom,
# and define our own operation instead (say, RECOG_PseuoRandom)?!
# Or at least change pseudorandomfunc to an attribute, so that
# the filter of this method can be turned into something like
# IsGroup and HasPseudoRandomFunc
if IsBound(g!.pseudorandomfunc) and Length(g!.pseudorandomfunc) > 0 then
l := Length(g!.pseudorandomfunc);
return CallFuncList(g!.pseudorandomfunc[l].func,
g!.pseudorandomfunc[l].args);
fi;
TryNextMethod();
end );See also #129, #237
Roughly the idea there is to re-use random elements for different purposes: if a random element was generated and used by naming, there is no reason why later on it should not be used again for constructive recognition. To handle this, recog maintains a stream of random elements, plus a bunch of "pointers" or "indices" into this stream, for different purposes. The challenge is to properly document this, and then use it.
Some more details are in this code comment:
Once all this is done, we could perhaps also get rid of the
PseudoRandomhack we have in place to get existing code to at least somewhat use these helpers:See also #129, #237