| This object holds all information needed to represent
the lookahead for any particular lookahead computation
for a single lookahead depth. Final lookahead
information is a simple bit set, but intermediate
stages need computation cycle and FOLLOW information.
Concerning the cycle variable.
If lookahead is computed for a RuleEnd node, then
computation is part of a FOLLOW cycle for this rule.
If lookahead is computed for a RuleBlock node, the
computation is part of a FIRST cycle to this rule.
Concerning the epsilonDepth variable.
This is not the depth relative to the rule reference
that epsilon was encountered. That value is
initial_k - epsilonDepth + 1
Also, lookahead depths past rule ref for local follow are:
initial_k - (initial_k - epsilonDepth)
Used for rule references. If we try
to compute look(k, ruleref) and there are fewer
than k lookahead terminals before the end of the
the rule, epsilon will be returned (don't want to
pass the end of the rule). We must track when the
the lookahead got stuck. For example,
a : b A B E F G;
b : C ;
LOOK(5, ref-to(b)) is {} with depth = 4, which
indicates that at 2 (5-4+1) tokens ahead, end of rule was reached.
Therefore, the token at 4=5-(5-4) past rule ref b must be
included in the set == F.
The situation is complicated by the fact that a computation
may hit the end of a rule at many different depths. For example,
a : b A B C ;
b : E F // epsilon depth of 1 relative to initial k=3
| G // epsilon depth of 2
;
Here, LOOK(3,ref-to(b)) returns epsilon, but the depths are
{1, 2}; i.e., 3-(3-1) and 3-(3-2). Those are the lookahead depths
past the rule ref needed for the local follow.
This is null unless an epsilon is created.
See Also: persistence.antlr.Lookahead.combineWith(Lookahead) |
