earlpy

1 files changed, 57 insertions, 0 deletions

diff --git a/DESIGN.txt b/DESIGN.txt
index 0762ca6..0fafff0 100644
--- a/DESIGN.txt
+++ b/DESIGN.txt
@@ -1,3 +1,60 @@
+====== Performance Issues ========
+The biggest performance issues involve parsing grammar
+
+    LIST nonterm
+        LIST , EXPR
+        EXPR
+
+With a huge input of
+
+    A , A , A , A , ...
+
+Naively, the parser will try to match an EXPR starting at each of the As.
+
+I guess this is because even if you are associating so far like
+
+    (((A, A), A), A), ...
+
+It doesn't know if, later on in the parse, you might want to reassociate 'across
+the boundary':
+
+    (((A, A), A), (A, B)), ...
+
+At this point, I don't think it's worth trying to do much more general-purpose
+optimization for this case. Instead, I think we should have callouts to user
+code to give hints. Two possible hint types:
+
+    1. "Parse this region as this parse tree." Easy to use: when you get there,
+       just skip all of those indices and complete that tree with any watchers.
+       (Might need to do prediction at the first index.)
+
+Actually, that's probably by far the easiest.
+
+====== More Disambiguation Issues ========
+What if we poison X, use it in completion Y, but then overwrite X with a
+nonpoisoned Z? Then Y will be incorrectly poisoned...
+
+====== More Disambiguation Issues ========
+Consider
+
+    STMTS nonterm .start
+        STMTS STMT
+        STMT
+
+vs.
+
+    STMTS nonterm .start
+        STMT
+        STMTS STMT
+
+Swapping this can actually impact what matches happen in STMT :O
+
+====== More Disambiguation Issues ========
+Associativity and precedence can have weird interplay
+
+E.g., maybe you can get Stmts(Error(1), Stmt(2, 3)) which has better
+associativity than Stmts(Stmt(1, 2), Stmt(3))
+
 ====== Disambiguation Issues ========
 Consider two possible parses of 0 + 1 * 5 + 4
     ( ( 0 + ( 1 *   5 ) ) + 4     )