1 _____________________________________________________________________________
4 - make sure I'm achieving optimal sharing (zipping heads and tails)
6 - clean up util package
8 - serializable parse tables?
9 - all that is required now is to separate Pos and Position
11 - currently we GC the doomed stack when the parent dies... but we
12 should also GC the parent when the doomed stack dies if it was a
15 - single-tree-return assumption
16 - have a way to let question marks not be tagged?
17 - "flat" sequences (no subtrees contain "::"s?) -- stringifiable
18 - make it so that we can have multi-result nonterminals
19 so long as they always appear under double-colons?
20 auto-insert the unwrap?
22 - get rid of Sequence.Singleton if possible
24 - use 'a'-'z' or 'a-z' instead of [a-z]?
27 foo.add(y.andnot(x)) ==> this is broken
28 - distinguish Conjunct from Sequence?
29 => !(Conjunct instanceof Reducible)
30 - avoid building the parts of the tree that end up getting dropped
31 - is it worth adding an additional class of states for these?
32 - or perhaps just a runtime node marker (hasNonDroppedParent)
33 - "ambiguity modulo dropped fragments"?
34 - this may conceal highly inefficient grammars...
35 - double-check all the region logic
36 - automatically collect time statistics and display
38 ______________________________________________________________________________
41 - MUST HAVE BETTER ERROR MESSAGES
42 - use for developing java15.g
43 - better ambiguity reporting
44 - colorized tree-diffs?
46 - better toString() methods all around...
48 - Treewalker code compiler?
49 - detect and reject circular gramars
51 - precedes restrictions ("<-")
52 - More topology untangling [later]
53 - grammar highlighting?
54 - Forest needs a "manual access" API (lifting makes this hard)
55 - rewriting language? multiple passes?
57 - Java grammar (java15.g)
59 - RFC2822 (email message/headers)
60 - Wikipedia grammar (needs to be both lexerless and boolean)
63 ______________________________________________________________________________
66 - try harder to "fuse" states together along two dimensions:
67 - identical (equivalent) states, or states that subsume each other
68 - unnecessary intermediate states ("short cut" GLR)
71 - better error messages
72 - Rekers & Koorn note that this can do really elegant and generalized "autocompletion".
74 - "Regular Right Part" grammars (NP Chapman, etc)
75 - Attribute unification
76 - Partly-Linear-PATR? (O(n^6) unification grammar)
78 - optional "prefer whitespace higher up" disambiguation heuristic
80 - Incremental parse table construction
82 - "lazy GLR" and "lazy trees" -> language with first-class CF matching
83 - perhaps linear boolean grammars instead? (linear time, quad space)
85 - Followed-by and not-followed-by predicates of arbitrary length
86 - expands the grammar beyond Boolean LR...
87 - requires *very* smart garbage collection
89 ______________________________________________________________________________
92 - understand and implement the RNGLR "kernel state" optimization.
93 The _Practical Early Parsing_ paper may help.
95 - implement Johnstone's algorithm for "reduced, resolved LR
96 tables" to eliminate superfluous reductions on
99 - Implement a k-token peek buffer (for each state, see if it "dead
100 ends" during the next k Phases based solely on state -- ignoring
103 - Is there any way we can avoid creating a GSS.Node instance for
104 nodes which are transient in the sense that they have only one
107 - Re-read Rekers, particularly the stuff on optimal sharing
109 - bring back in parse-table phase resolution of precedence (just
110 like associativity). This can be inferred from the use of ">"
111 when the rules are in one of these special forms:
122 where "_" is anything and "E" is the defining nonterminal.
123 Essentially what we're looking for is the situation where the
124 leftmost portion of one rule produces another rule, and the
125 rightmost portion of the latter produces the former.
127 I'm not 100% certain that this is as "strong" as the prefer/avoid
128 form (try to prove this, you probably can), but it's "what people
129 intend" most of the time.