Faster regular expression pattern matching in the grammar compiler.

The sequence operator (x+y) was implemented by splitting the string to be
matched at all positions and trying to match the parts against the two
subpatterns. To reduce the number of splits, we now estimate the minimum and
maximum length of the string that the subpatterns could match. For common
cases, where one of the subpatterns is a string of known length, like
in (x+"y") or (x + ("a"|"o"|"u"|"e")+"y"), only one split will be tried.

src/compiler/GF/Grammar/Grammar.hs

@@ -430,6 +430,7 @@ data Term =
  | Error String                  -- ^ error values returned by Predef.error
   deriving (Show, Eq, Ord)
 
+-- | Patterns
 data Patt =
    PC Ident [Patt]        -- ^ constructor pattern: @C p1 ... pn@    @C@ 
  | PP QIdent [Patt]       -- ^ package constructor pattern: @P.C p1 ... pn@    @P.C@ 
@@ -450,14 +451,17 @@ data Patt =
  | PNeg Patt              -- ^ negated pattern: -p
  | PAlt Patt Patt         -- ^ disjunctive pattern: p1 | p2
  | PSeq Patt Patt         -- ^ sequence of token parts: p + q
+ | PMSeq MPatt MPatt      -- ^ sequence of token parts: p + q
  | PRep Patt              -- ^ repetition of token part: p*
  | PChar                  -- ^ string of length one: ?
  | PChars [Char]          -- ^ character list: ["aeiou"]
  | PMacro Ident           -- #p
  | PM QIdent              -- #m.p
-
   deriving (Show, Eq, Ord)
 
+-- | Measured pattern (paired with the min & max matching length)
+type MPatt = ((Int,Int),Patt)
+
 -- | to guide computation and type checking of tables
 data TInfo = 
    TRaw         -- ^ received from parser; can be anything