fredokun / arbogen Goto Github PK

The following spec works:

set min 100;
set max 120;
set try 500;

Functor ::= Term * Term ;
CFunctor ::= Term * Term ;
Term ::= Name * + Functor * + CFunctor * ;

however it generates 'Term' nodes whereas 'Term' is not a directly recursive rule

If we change the same spec as follows:

set min 100;
set max 120;
set try 5000;

Term ::= Name * + Functor * Term * Term * + CFunctor * Term * Term * ;

then only trees of size 1 are generated

Finally, with:

set min 100;
set max 120;
set try 5000;

Term ::= Name * + Functor * Term * Term * * + CFunctor * Term * Term * ;

There seems to be an infinite loop

If the user sets a min that is bigger then the max in his grammar , Arbogen doesn't detect the problem and keeps running trying to find the tree.

Add the possibility to sample several trees with same parameters

Is it worth doing indentation in dot files ?
It causes performance issue :
for example, with examples/binary100000.spec settings and seed=4242,
it takes 14 seconds on my machine to generate and export the tree to non-indented dot file, vs. almost 3 minutes with indentation.

Another spec that should actually work (was proposed by Mr Antoine)

set zstart 0.1;
set min 10;
set max 2000;
set try 500;

Tree ::= Leaf * * + Tree * SEQ(Tree) * ;

Pour le packaging Debian j'ai un idée , pourquoi pas notre script de packaging soit global et pas pour un architecture specific. Ca vaut dire quand on lance le script il détecte l'architecture d'ordinateur et tous cas il va compile après donc on est sure notre exécutable va marcher. Ca nous fait un script pour tous et pas plusier.

We currently try to do singular sampling regardless of the grammar at play.

This approach is numerically unstable and fails on linear grammars.

It seems better to tune the value of the parameter z so that the expected size of the output matches size_max