1 // priorities are all messy and dont get serialized
3 // 2. Java MetaGrammar (proof of concept)
4 // 3. Ivan's MetaGrammar
5 // 4. Documentation format
8 // TODO: better API for interfacing with Java
9 // TODO: error messages
10 // TODO: integrate with TIB
15 // MetaGrammar (necessary/relevant?)
16 // Tree<String> (cleanup?)
20 // FEATURE: serialization of ParseTable's, generation of Java code
21 // FEATURE: infer reject elements for literals
22 // FEATURE: prefer whitespace higher up
23 // FEATURE: full conjunctive and boolean grammars
24 // FEATURE: "ambiguity modulo dropped fragments"? can this be checked for statically? eliminated statically?
25 // - drop stuff during the parsing process (drop nodes)
27 // LATER: Element<A> -- parameterize over the input token type? Makes a huge mess...
28 // LATER: Go back to where Sequence is not an Element?
29 // - The original motivation for making Sequence "first class" was the fact that
30 // in order to do associativity right you need to have per-Sequence follow sets
32 ______________________________________________________________________________
35 - switch maximal to not-followed-by (~/~)
37 - should Union.add() be there?
38 - should Atom.top() be there?
40 - fix the location stuff, it's broken
41 - decent/better error messages
47 - RFC2822 (email message/headers)
50 ______________________________________________________________________________
53 - clean up the whole Walk situation
55 - cleaner solution to "maximal"?
58 - right now I can only lift the last child in a forest... begs
59 the question of what the right representation for Forests is
60 if we need to be able to do lift operations on it.
63 - "Regular Right Part" grammars (NP Chapman, etc)
64 - Attribute unification
66 ______________________________________________________________________________
69 - Implement a k-token peek buffer (for each state, see if it "dead
70 ends" during the next k Phases based solely on state -- ignoring
73 - Arrange for the SPPF corresponding to dropped subtrees to never be
74 generated (or merged, etc)
76 - Is there any way we can avoid creating a GSS.Node instance for
77 nodes which are transient in the sense that they have only one
80 - Implement "GLR syntactic predicates" -- the ability to do
81 arbitrary lookahead (ie "followed-by" and "not-followed-by" for
82 arbitrary patterns). This enables generalized longest-match and
83 lets us drop the Maximal hack.
85 - Re-read Rekers, particularly the stuff on optimal sharing
87 - Isolate the Element objects from Parse.Table/GSS so we can move
90 - consider allowing a Forest.Body to represent some other Tree whose
91 Body's should be [recursively] considered part of this Forest.
93 - perhaps not: right now we have a nice situation where
94 Forest.Ref instances become immutable once iterator()ed. This
95 also gives us a strong place to to culling with the certainty
96 that we won't throw out a Body which would later be salvaged
97 by some yet-to-be-added dependency.
99 - Figure out if there is a way to:
101 - allow unwrapping of children other than the very last one.
103 - fold repetitions into an array form in Forest, before
104 conversion to Tree. The major problem here is that multiple
105 tree-arrays are possible, all of different lengths. Worse,
106 even if they're all the same length, not all elements belong
107 in the same "possibility vector" as all others. You
108 essentially need a GSS to represent the array, which perhaps
109 is what the unfolded form was in the first place.
111 - Wikipedia grammar (needs to be both lexerless and boolean)
114 => Ordered Choice (";" operator)
116 - bring back in parse-table phase resolution of precedence (just
117 like associativity). This can be inferred from the use of ">"
118 when the rules are in one of these special forms:
129 where "_" is anything and "E" is the defining nonterminal.
130 Essentially what we're looking for is the situation where the
131 leftmost portion of one rule produces another rule, and the
132 rightmost portion of the latter produces the former.
134 I'm not 100% certain that this is as "strong" as the prefer/avoid
135 form (try to prove this, you probably can), but it's "what people
136 intend" most of the time.
138 - implement Johnstone's algorithm for "reduced, resolved LR
139 tables" to eliminate superfluous reductions on