Incremental LR Parsing

Given a parse tree for a sentence xyz and a string w , an incremental parser builds the parse tree for the modified sentence xwz by re-using as much of the parse tree for xyz as possible. A key observation of the LR parser is that, if a string of tokens α can be reduced to a nonterminal A in some state, then α must be reduced to A in all "similar" states as long as the look-ahead condition of the states is satisfied. There is no other way to parse α. Therefore, the subtree that derives α can be re-used directly. We present an incremental LR parsing algorithm based on the observation. The incremental parser is able to divide its work among parallel processes. The question whether two states are "similar" can be answered by checking the goto table in the conventional LR parsers.