improve perf

Author: nitely

src/nregex.nim

@@ -241,8 +241,10 @@ func find*(
   var i = start
   var c: Rune
   while i < len(s):
-    result = matchImpl(s, pattern, m, {mfLongestMatch}, i)
+    result = matchImpl(s, pattern, m, {mfShortestMatch, mfNoCaptures}, i)
     if result:
+      result = matchImpl(s, pattern, m, {mfLongestMatch}, i)
+      doAssert result
       break
     fastRuneAt(s, i, c, true)
 

src/nregex/dfamacro.nim

@@ -303,13 +303,21 @@ func matchImpl*(
     qt = q
     i = start
     iPrev = start
+  # workaround for VM registry limitation
+  const
+    tzLen = regex.transitions.z.len
+    groupTransitionsLen = regex.groupsCount * 2
+    noCaptures = mfNoCaptures in flags
   # workaround for https://github.com/nim-lang/Nim/issues/13252
   const
     reFlags = regex.flags
-    hasTransionsZ = regex.transitions.z.len > 0
+    canSkipTransitionsZ = noCaptures and
+      groupTransitionsLen == tzLen
+    hasTransitionsZ = tzLen > 0 and
+      not canSkipTransitionsZ
     groupCount {.used.} = regex.groupsCount
     namedGroups {.used.} = regex.namedGroups
-  when hasTransionsZ:
+  when hasTransitionsZ:
     var
       smA = newSubmatches()
       smB = newSubmatches()
@@ -329,12 +337,12 @@ func matchImpl*(
         genSymMatchTable(q, qt, c.int32, regex)
       if (q == -1'i32).unlikely:
         return
-    when hasTransionsZ:
+    when hasTransitionsZ:
       submatch(smA, smB, capts, regex, iPrev, qt, cPrev, c.int32)
     iPrev = i
     cPrev = c.int32
   genEoeTable(result, q, qt, regex)
-  when hasTransionsZ:
+  when hasTransitionsZ:
     if not result:
       return
     # XXX lighter submatchEoe

src/nregex/dfamatch.nim

@@ -200,7 +200,7 @@ template longestMatchEnter(): untyped {.dirty.} =
   if symEoe in regex.dfa.table[q]:
     matchedLong = true
     iPrevLong = iPrev
-    if regex.transitions.z.len > 0:
+    if hasTransitionsZ:
       submatch(
         smA, smB, capts, regex.transitions,
         regex.dfa.cs[regex.dfa.closures[q][symEoe]], iPrev, cPrev, c.int32)
@@ -210,7 +210,7 @@ template longestMatchEnter(): untyped {.dirty.} =
 
 template longestMatchExit(): untyped {.dirty.} =
   result = matchedLong
-  if regex.transitions.z.len > 0:
+  if hasTransitionsZ:
     constructSubmatches(m.captures, capts, captLong, regex.groupsCount)
     if regex.namedGroups.len > 0:
       m.namedGroups = regex.namedGroups
@@ -219,7 +219,7 @@ template longestMatchExit(): untyped {.dirty.} =
 
 template shortestMatch(): untyped {.dirty.} =
   if symEoe in regex.dfa.table[q]:
-    if regex.transitions.z.len > 0:
+    if hasTransitionsZ:
       submatch(
         smA, smB, capts, regex.transitions,
         regex.dfa.cs[regex.dfa.closures[q][symEoe]], iPrev, cPrev, c.int32)
@@ -241,7 +241,12 @@ func matchImpl*(
   #echo dfa
   m.clear()
   result = false
-  let asciiMode = reAscii in regex.flags
+  let
+    asciiMode = reAscii in regex.flags
+    canSkipTransitionsZ = mfNoCaptures in flags and
+      regex.groupsCount * 2 == regex.transitions.z.len
+    hasTransitionsZ = regex.transitions.z.len > 0 and
+      not canSkipTransitionsZ
   var
     smA: Submatches
     smB: Submatches
@@ -257,7 +262,7 @@ func matchImpl*(
     matchedLong {.used.} = false
     captLong {.used.} = -1
     iPrevLong {.used.} = start
-  if regex.transitions.z.len > 0:
+  if hasTransitionsZ:
     smA = newSubmatches()
     smB = newSubmatches()
     smA.add((0'i16, -1'i32))
@@ -283,7 +288,7 @@ func matchImpl*(
           longestMatchExit()
         else:
           return
-    if regex.transitions.z.len > 0:
+    if hasTransitionsZ:
       submatch(
         smA, smB, capts, regex.transitions,
         regex.dfa.cs[regex.dfa.closures[q][cSym]], iPrev, cPrev, c.int32)
@@ -296,7 +301,7 @@ func matchImpl*(
     when mfLongestMatch in flags:
       longestMatchExit()
     return
-  if regex.transitions.z.len > 0:
+  if hasTransitionsZ:
     submatch(
       smA, smB, capts, regex.transitions,
       regex.dfa.cs[regex.dfa.closures[q][symEoe]], iPrev, cPrev, -1'i32)

src/nregex/nfa.nim

@@ -209,6 +209,7 @@ type
     allZ*: TransitionsAll
     z*: ZclosureStates
 
+# XXX do not add char classes transitions \w, \d, etc in ascii mode
 func eRemoval(
   eNfa: seq[Node],
   transitions: var Transitions