SDFG class documentation does not match implementation

[tim0s]

The SDFG class init method looks like this:

def __init__(self,
                 name: str,
                 constants: Dict[str, Tuple[dt.Data, Any]] = None,
                 propagate: bool = True,
                 parent=None):
        """ Constructs a new SDFG.

            :param name: Name for the SDFG (also used as the filename for
                         the compiled shared library).
            :param symbols: Additional dictionary of symbol names -> types that the SDFG
                            defines, apart from symbolic data sizes.
            :param propagate: If False, disables automatic propagation of
                              memlet subsets from scopes outwards. Saves
                              processing time but disallows certain
                              transformations.
            :param parent: The parent SDFG or SDFG state (for nested SDFGs).
        """

Note that the second argument is named constants in the implementation but symbols in the documentation.
I don't want to just change it because I don't fully understand what is meant here and for me the semantics are different: Constants do not change within their scope of definition, Symbols (as an SDFG concept) can be changed i.e., on interstate edges AFAIU.

[alexnick83]

The documentation is wrong. It should be constants. Symbols are added to SDFGs as follows:

• Explicitly: call SDFG.add_symbol (please note that direct manipulation of the SDFG.symbols dictionary is also allowed)
• Implicitly: whenever you add any elements on the SDFG that depend on/refer to specific symbols. For example, in the following code snippet, N will be implicitly added to the SDFG and the output should be {'N': int}:

import dace
N = dace.symbol('N')
sdfg = dace.SDFG('implicit_symbol_test')
sdfg.add_array('A', shape=(N,), dtype=dace.int32)
print(sdfg.symbols)

I will make a PR.

[tim0s]

Fixed by #1571