diff --git a/config/nimrod.cfg b/config/nimrod.cfg
index 22dc35f5af..122980348c 100755
--- a/config/nimrod.cfg
+++ b/config/nimrod.cfg
@@ -57,6 +57,7 @@ hint[LineTooLong]=off
 @if unix:
   @if not bsd:
     gcc.options.linker = "-ldl"
+    tcc.options.linker = "-ldl"
   @end
 @end
 
diff --git a/doc/manual.txt b/doc/manual.txt
index 9a60602641..5038e292b5 100755
--- a/doc/manual.txt
+++ b/doc/manual.txt
@@ -2542,10 +2542,10 @@ cannot be inherited from.
 Pure pragma
 -----------
 The `pure`:idx: pragma serves two completely different purposes:
-1) To mark a procedure that Nimrod should not generate any exit statements like
+1. To mark a procedure that Nimrod should not generate any exit statements like
    ``return result;`` in the generated code. This is useful for procs that only
    consist of an assembler statement.
-2) To mark an object type so that its type field should be omitted. This is
+2. To mark an object type so that its type field should be omitted. This is
    necessary for binary compatibility with other compiled languages.
 
 
@@ -2795,7 +2795,7 @@ a dynamic library. The pragma then has no argument and has to be used in
 conjunction with the ``exportc`` pragma:
 
 .. code-block:: Nimrod
-  proc exportme(): int {.cdecl, export, dynlib.}
+  proc exportme(): int {.cdecl, exportc, dynlib.}
 
 This is only useful if the program is compiled as a dynamic library via the
 ``--app:lib`` command line option.
diff --git a/doc/pegdocs.txt b/doc/pegdocs.txt
index 922b998c74..6dde731113 100755
--- a/doc/pegdocs.txt
+++ b/doc/pegdocs.txt
@@ -98,6 +98,11 @@ macro              meaning
 ``\i``             ignore case for matching; use this at the start of the PEG
 ``\y``             ignore style for matching; use this at the start of the PEG
 ``\ident``         a standard ASCII identifier: ``[a-zA-Z_][a-zA-Z_0-9]*``
+``\letter``        any Unicode letter
+``\upper``         any Unicode uppercase letter
+``\lower``         any Unicode lowercase letter
+``\title``         any Unicode title letter
+``\white``         any Unicode whitespace character
 ==============     ============================================================
 
 A backslash followed by a letter is a built-in macro, otherwise it
diff --git a/doc/tut2.txt b/doc/tut2.txt
index a139fb5deb..bd9769af31 100755
--- a/doc/tut2.txt
+++ b/doc/tut2.txt
@@ -75,9 +75,9 @@ Inheritance is done with the ``object of`` syntax. Multiple inheritance is
 currently not supported. If an object type has no suitable ancestor, ``TObject``
 can be used as its ancestor, but this is only a convention.
 
-**Note**: Aggregation (*has-a* relation) is often preferable to inheritance
+**Note**: Composition (*has-a* relation) is often preferable to inheritance
 (*is-a* relation) for simple code reuse. Since objects are value types in
-Nimrod, aggregation is as efficient as inheritance.
+Nimrod, composition is as efficient as inheritance.
 
 
 Mutually recursive types
@@ -487,7 +487,7 @@ The example shows a generic binary tree. Depending on context, the brackets are
 used either to introduce type parameters or to instantiate a generic proc,
 iterator or type. As the example shows, generics work with overloading: the
 best match of ``add`` is used. The built-in ``add`` procedure for sequences
-is not hidden and used in the ``preorder`` iterator.
+is not hidden and is used in the ``preorder`` iterator.
 
 
 Templates
diff --git a/examples/httpserver2.nim b/examples/httpserver2.nim
new file mode 100644
index 0000000000..0604e6a83c
--- /dev/null
+++ b/examples/httpserver2.nim
@@ -0,0 +1,245 @@
+import strutils, os, osproc, strtabs, streams, sockets
+
+const
+  wwwNL* = "\r\L"
+  ServerSig = "Server: httpserver.nim/1.0.0" & wwwNL
+
+type
+  TRequestMethod = enum reqGet, reqPost
+  TServer* = object       ## contains the current server state
+    socket: TSocket
+    job: seq[TJob]
+  TJob* = object
+    client: TSocket
+    process: PProcess
+
+# --------------- output messages --------------------------------------------
+
+proc sendTextContentType(client: TSocket) =
+  send(client, "Content-type: text/html" & wwwNL)
+  send(client, wwwNL)
+
+proc badRequest(client: TSocket) =
+  # Inform the client that a request it has made has a problem.
+  send(client, "HTTP/1.0 400 BAD REQUEST" & wwwNL)
+  sendTextContentType(client)
+  send(client, "<p>Your browser sent a bad request, " &
+               "such as a POST without a Content-Length.</p>" & wwwNL)
+
+
+proc cannotExec(client: TSocket) =
+  send(client, "HTTP/1.0 500 Internal Server Error" & wwwNL)
+  sendTextContentType(client)
+  send(client, "<P>Error prohibited CGI execution.</p>" & wwwNL)
+
+
+proc headers(client: TSocket, filename: string) =
+  # XXX could use filename to determine file type
+  send(client, "HTTP/1.0 200 OK" & wwwNL)
+  send(client, ServerSig)
+  sendTextContentType(client)
+
+proc notFound(client: TSocket, path: string) =
+  send(client, "HTTP/1.0 404 NOT FOUND" & wwwNL)
+  send(client, ServerSig)
+  sendTextContentType(client)
+  send(client, "<html><title>Not Found</title>" & wwwNL)
+  send(client, "<body><p>The server could not fulfill" & wwwNL)
+  send(client, "your request because the resource <b>" & path & "</b>" & wwwNL)
+  send(client, "is unavailable or nonexistent.</p>" & wwwNL)
+  send(client, "</body></html>" & wwwNL)
+
+
+proc unimplemented(client: TSocket) =
+  send(client, "HTTP/1.0 501 Method Not Implemented" & wwwNL)
+  send(client, ServerSig)
+  sendTextContentType(client)
+  send(client, "<html><head><title>Method Not Implemented" &
+               "</title></head>" &
+               "<body><p>HTTP request method not supported.</p>" &
+               "</body></HTML>" & wwwNL)
+
+
+# ----------------- file serving ---------------------------------------------
+
+proc discardHeaders(client: TSocket) = skip(client)
+
+proc serveFile(client: TSocket, filename: string) =
+  discardHeaders(client)
+
+  var f: TFile
+  if open(f, filename):
+    headers(client, filename)
+    const bufSize = 8000 # != 8K might be good for memory manager
+    var buf = alloc(bufsize)
+    while True:
+      var bytesread = readBuffer(f, buf, bufsize)
+      if bytesread > 0:
+        var byteswritten = send(client, buf, bytesread)
+        if bytesread != bytesWritten:
+          dealloc(buf)
+          close(f)
+          OSError()
+      if bytesread != bufSize: break
+    dealloc(buf)
+    close(f)
+    client.close()
+  else:
+    notFound(client, filename)
+
+# ------------------ CGI execution -------------------------------------------
+
+proc executeCgi(server: var TServer, client: TSocket, path, query: string, 
+                meth: TRequestMethod) =
+  var env = newStringTable(modeCaseInsensitive)
+  var contentLength = -1
+  case meth
+  of reqGet:
+    discardHeaders(client)
+
+    env["REQUEST_METHOD"] = "GET"
+    env["QUERY_STRING"] = query
+  of reqPost:
+    var buf = ""
+    var dataAvail = true
+    while dataAvail:
+      dataAvail = recvLine(client, buf)
+      if buf.len == 0:
+        break
+      var L = toLower(buf)
+      if L.startsWith("content-length:"):
+        var i = len("content-length:")
+        while L[i] in Whitespace: inc(i)
+        contentLength = parseInt(copy(L, i))
+
+    if contentLength < 0:
+      badRequest(client)
+      return
+
+    env["REQUEST_METHOD"] = "POST"
+    env["CONTENT_LENGTH"] = $contentLength
+
+  send(client, "HTTP/1.0 200 OK" & wwwNL)
+
+  var process = startProcess(command=path, env=env)
+ 
+  var job: TJob
+  job.process = process
+  job.client = client
+  server.job.add(job)
+ 
+  if meth == reqPost:
+    # get from client and post to CGI program:
+    var buf = alloc(contentLength)
+    if recv(client, buf, contentLength) != contentLength: 
+      dealloc(buf)
+      OSError()
+    var inp = process.inputStream
+    inp.writeData(inp, buf, contentLength)
+    dealloc(buf)
+
+proc animate(server: var TServer) =
+  # checks list of jobs, removes finished ones (pretty sloppy by seq copying)
+  var active_jobs: seq[TJob] = @[]
+  for i in 0..server.job.len-1:
+    var job = server.job[i]
+    if running(job.process):
+      active_jobs.add(job)
+    else:
+      # read process output stream and send it to client
+      var outp = job.process.outputStream
+      while true:
+        var line = outp.readstr(1024)
+        if line.len == 0:
+          break
+        else:
+          try:
+            send(job.client, line)
+          except:
+            echo("send failed, client diconnected")
+      close(job.client)
+
+  server.job = active_jobs
+
+# --------------- Server Setup -----------------------------------------------
+
+proc acceptRequest(server: var TServer, client: TSocket) =
+  var cgi = false
+  var query = ""
+  var buf = ""
+  discard recvLine(client, buf)
+  var path = ""
+  var data = buf.split()
+  var meth = reqGet
+  var q = find(data[1], '?')
+
+  # extract path
+  if q >= 0:
+    # strip "?..." from path, this may be found in both POST and GET
+    path = data[1].copy(0, q-1)
+  else:
+    path = data[1]
+  # path starts with "/", by adding "." in front of it we serve files from cwd
+  path = "." & path
+
+  echo("accept: " & path)
+
+  if cmpIgnoreCase(data[0], "GET") == 0:
+    if q >= 0:
+      cgi = true
+      query = data[1].copy(q+1)
+  elif cmpIgnoreCase(data[0], "POST") == 0:
+    cgi = true
+    meth = reqPost
+  else:
+    unimplemented(client)
+
+  if path[path.len-1] == '/' or existsDir(path):
+    path = path / "index.html"
+
+  if not ExistsFile(path):
+    discardHeaders(client)
+    notFound(client, path)
+    client.close()
+  else:
+    when defined(Windows):
+      var ext = splitFile(path).ext.toLower
+      if ext == ".exe" or ext == ".cgi":
+        # XXX: extract interpreter information here?
+        cgi = true
+    else:
+      if {fpUserExec, fpGroupExec, fpOthersExec} * path.getFilePermissions != {}:
+        cgi = true
+    if not cgi:
+      serveFile(client, path)
+    else:
+      executeCgi(server, client, path, query, meth)
+
+
+
+when isMainModule:
+  var port = 80
+
+  var server: TServer
+  server.job = @[]
+  server.socket = socket(AF_INET)
+  if server.socket == InvalidSocket: OSError()
+  server.socket.bindAddr(port=TPort(port))
+  listen(server.socket)
+  echo("server up on port " & $port)
+
+  while true:
+    # check for new new connection & handle it
+    var list: seq[TSocket] = @[server.socket]
+    if select(list, 10) > 0:
+      var client = accept(server.socket)
+      try:
+        acceptRequest(server, client)
+      except:
+        echo("failed to accept client request")
+
+    # pooling events
+    animate(server)
+    # some slack for CPU
+    sleep(10)
+  server.socket.close()
diff --git a/lib/nimbase.h b/lib/nimbase.h
index 983bb112dc..0251364d14 100755
--- a/lib/nimbase.h
+++ b/lib/nimbase.h
@@ -22,6 +22,10 @@ __TINYC__
 #ifndef NIMBASE_H
 #define NIMBASE_H
 
+#if defined(__GNUC__)
+#  define _GNU_SOURCE 1
+#endif
+
 #if !defined(__TINYC__)
 #  include  <math.h>
 #else
@@ -29,7 +33,6 @@ __TINYC__
 #  define GCC_MAJOR 4
 #  define __GNUC_MINOR__ 4
 #  define __GNUC_PATCHLEVEL__ 5 */
-
 #  define __DECLSPEC_SUPPORTED 1
 #endif
 
diff --git a/lib/pure/cgi.nim b/lib/pure/cgi.nim
index 643c67bec8..af222cabaa 100755
--- a/lib/pure/cgi.nim
+++ b/lib/pure/cgi.nim
@@ -7,7 +7,7 @@
 #    distribution, for details about the copyright.
 #
 
-## This module implements helper procs for CGI applictions. Example:
+## This module implements helper procs for CGI applications. Example:
 ## 
 ## .. code-block:: Nimrod
 ##
@@ -29,7 +29,7 @@
 ##    writeln(stdout, "your password: " & myData["password"])
 ##    writeln(stdout, "</body></html>")
 
-import strutils, os, strtabs
+import strutils, os, strtabs, cookies
 
 proc URLencode*(s: string): string =
   ## Encodes a value to be HTTP safe: This means that characters in the set
@@ -355,32 +355,16 @@ proc setCookie*(name, value: string) =
   write(stdout, "Set-Cookie: ", name, "=", value, "\n")
 
 var
-  cookies: PStringTable = nil
-
-proc parseCookies(s: string): PStringTable = 
-  result = newStringTable(modeCaseInsensitive)
-  var i = 0
-  while true:
-    while s[i] == ' ' or s[i] == '\t': inc(i)
-    var keystart = i
-    while s[i] != '=' and s[i] != '\0': inc(i)
-    var keyend = i-1
-    if s[i] == '\0': break
-    inc(i) # skip '='
-    var valstart = i
-    while s[i] != ';' and s[i] != '\0': inc(i)
-    result[copy(s, keystart, keyend)] = copy(s, valstart, i-1)
-    if s[i] == '\0': break
-    inc(i) # skip ';'
+  gcookies: PStringTable = nil
     
 proc getCookie*(name: string): string = 
   ## Gets a cookie. If no cookie of `name` exists, "" is returned.
-  if cookies == nil: cookies = parseCookies(getHttpCookie())
-  result = cookies[name]
+  if gcookies == nil: gcookies = parseCookies(getHttpCookie())
+  result = gcookies[name]
 
 proc existsCookie*(name: string): bool = 
   ## Checks if a cookie of `name` exists.
-  if cookies == nil: cookies = parseCookies(getHttpCookie())
-  result = hasKey(cookies, name)
+  if gcookies == nil: gcookies = parseCookies(getHttpCookie())
+  result = hasKey(gcookies, name)
 
 
diff --git a/lib/pure/cookies.nim b/lib/pure/cookies.nim
new file mode 100644
index 0000000000..eed6c7512a
--- /dev/null
+++ b/lib/pure/cookies.nim
@@ -0,0 +1,30 @@
+#
+#
+#            Nimrod's Runtime Library
+#        (c) Copyright 2010 Andreas Rumpf
+#
+#    See the file "copying.txt", included in this
+#    distribution, for details about the copyright.
+#
+
+## This module implements helper procs for parsing Cookies.
+
+import strtabs
+
+proc parseCookies*(s: string): PStringTable = 
+  ## parses cookies into a string table.
+  result = newStringTable(modeCaseInsensitive)
+  var i = 0
+  while true:
+    while s[i] == ' ' or s[i] == '\t': inc(i)
+    var keystart = i
+    while s[i] != '=' and s[i] != '\0': inc(i)
+    var keyend = i-1
+    if s[i] == '\0': break
+    inc(i) # skip '='
+    var valstart = i
+    while s[i] != ';' and s[i] != '\0': inc(i)
+    result[copy(s, keystart, keyend)] = copy(s, valstart, i-1)
+    if s[i] == '\0': break
+    inc(i) # skip ';'
+
diff --git a/lib/pure/httpserver.nim b/lib/pure/httpserver.nim
index bff1b381d4..cb36ea5417 100755
--- a/lib/pure/httpserver.nim
+++ b/lib/pure/httpserver.nim
@@ -136,9 +136,12 @@ proc executeCgi(client: TSocket, path, query: string, meth: TRequestMethod) =
   if meth == reqPost:
     # get from client and post to CGI program:
     var buf = alloc(contentLength)
-    if recv(client, buf, contentLength) != contentLength: OSError()
+    if recv(client, buf, contentLength) != contentLength: 
+      dealloc(buf)
+      OSError()
     var inp = process.inputStream
     inp.writeData(inp, buf, contentLength)
+    dealloc(buf)
 
   var outp = process.outputStream
   while running(process) or not outp.atEnd(outp):
@@ -153,10 +156,21 @@ proc acceptRequest(client: TSocket) =
   var query = ""
   var buf = ""
   discard recvLine(client, buf)
+  var path = ""
   var data = buf.split()
   var meth = reqGet
+
+  var q = find(data[1], '?')
+
+  # extract path
+  if q >= 0:
+    # strip "?..." from path, this may be found in both POST and GET
+    path = "." & data[1].copy(0, q-1)
+  else:
+    path = "." & data[1]
+  # path starts with "/", by adding "." in front of it we serve files from cwd
+  
   if cmpIgnoreCase(data[0], "GET") == 0:
-    var q = find(data[1], '?')
     if q >= 0:
       cgi = true
       query = data[1].copy(q+1)
@@ -166,7 +180,6 @@ proc acceptRequest(client: TSocket) =
   else:
     unimplemented(client)
 
-  var path = data[1]
   if path[path.len-1] == '/' or existsDir(path):
     path = path / "index.html"
 
@@ -221,7 +234,7 @@ proc next*(s: var TServer) =
   var buf = ""
   discard recvLine(s.client, buf)
   var data = buf.split()
-  if cmpIgnoreCase(data[0], "GET") == 0:
+  if cmpIgnoreCase(data[0], "GET") == 0 or cmpIgnoreCase(data[0], "POST") == 0:
     var q = find(data[1], '?')
     if q >= 0:
       s.query = data[1].copy(q+1)
diff --git a/lib/pure/pegs.nim b/lib/pure/pegs.nim
index 0a373125d9..175713a0c4 100755
--- a/lib/pure/pegs.nim
+++ b/lib/pure/pegs.nim
@@ -37,6 +37,11 @@ type
     pkAny,              ## any character (.)
     pkAnyRune,          ## any Unicode character (_)
     pkNewLine,          ## CR-LF, LF, CR
+    pkLetter,           ## Unicode letter
+    pkLower,            ## Unicode lower case letter
+    pkUpper,            ## Unicode upper case letter
+    pkTitle,            ## Unicode title character
+    pkWhitespace,       ## Unicode whitespace character
     pkTerminal,
     pkTerminalIgnoreCase,
     pkTerminalIgnoreStyle,
@@ -71,7 +76,7 @@ type
     rule: TNode                   ## the rule that the symbol refers to
   TNode {.final.} = object
     case kind: TPegKind
-    of pkEmpty, pkAny, pkAnyRune, pkGreedyAny, pkNewLine: nil
+    of pkEmpty..pkWhitespace: nil
     of pkTerminal, pkTerminalIgnoreCase, pkTerminalIgnoreStyle: term: string
     of pkChar, pkGreedyRepChar: ch: char
     of pkCharChoice, pkGreedyRepSet: charChoice: ref set[char]
@@ -196,6 +201,7 @@ proc `@`*(a: TPeg): TPeg {.nosideEffect, rtl, extern: "npegsSearch".} =
 
 proc `@@`*(a: TPeg): TPeg {.noSideEffect, rtl, 
                             extern: "npgegsCapturedSearch".} =
+  ## constructs a "captured search" for the PEG `a`
   result.kind = pkCapturedSearch
   result.sons = @[a]
   
@@ -237,6 +243,27 @@ proc newLine*: TPeg {.inline.} =
   ## constructs the PEG `newline`:idx: (``\n``)
   result.kind = pkNewline
 
+proc UnicodeLetter*: TPeg {.inline.} = 
+  ## constructs the PEG ``\letter`` which matches any Unicode letter.
+  result.kind = pkLetter
+  
+proc UnicodeLower*: TPeg {.inline.} = 
+  ## constructs the PEG ``\lower`` which matches any Unicode lowercase letter.
+  result.kind = pkLower 
+
+proc UnicodeUpper*: TPeg {.inline.} = 
+  ## constructs the PEG ``\upper`` which matches any Unicode lowercase letter.
+  result.kind = pkUpper 
+  
+proc UnicodeTitle*: TPeg {.inline.} = 
+  ## constructs the PEG ``\title`` which matches any Unicode title letter.
+  result.kind = pkTitle
+
+proc UnicodeWhitespace*: TPeg {.inline.} = 
+  ## constructs the PEG ``\white`` which matches any Unicode 
+  ## whitespace character.
+  result.kind = pkWhitespace
+
 proc capture*(a: TPeg): TPeg {.nosideEffect, rtl, extern: "npegsCapture".} =
   ## constructs a capture with the PEG `a`
   result.kind = pkCapture
@@ -267,8 +294,8 @@ proc spaceCost(n: TPeg): int =
   case n.kind
   of pkEmpty: nil
   of pkTerminal, pkTerminalIgnoreCase, pkTerminalIgnoreStyle, pkChar,
-     pkGreedyRepChar, pkCharChoice, pkGreedyRepSet, pkAny, pkAnyRune,
-     pkNewLine, pkGreedyAny:
+     pkGreedyRepChar, pkCharChoice, pkGreedyRepSet, 
+     pkAny..pkWhitespace, pkGreedyAny:
     result = 1
   of pkNonTerminal:
     # we cannot inline a rule with a non-terminal
@@ -379,6 +406,12 @@ proc toStrAux(r: TPeg, res: var string) =
   of pkEmpty: add(res, "()")
   of pkAny: add(res, '.')
   of pkAnyRune: add(res, '_')
+  of pkLetter: add(res, "\\letter")
+  of pkLower: add(res, "\\lower")
+  of pkUpper: add(res, "\\upper")
+  of pkTitle: add(res, "\\title")
+  of pkWhitespace: add(res, "\\white")
+
   of pkNewline: add(res, "\\n")
   of pkTerminal: add(res, singleQuoteEsc(r.term))
   of pkTerminalIgnoreCase:
@@ -460,10 +493,15 @@ proc `$` *(r: TPeg): string {.nosideEffect, rtl, extern: "npegsToString".} =
 # --------------------- core engine -------------------------------------------
 
 type
-  TMatchClosure {.final.} = object
+  TCaptures* {.final.} = object ## contains the captured substrings.
     matches: array[0..maxSubpatterns-1, tuple[first, last: int]]
     ml: int
 
+proc bounds*(c: TCaptures, 
+             i: range[0..maxSubpatterns-1]): tuple[first, last: int] = 
+  ## returns the bounds ``[first..last]`` of the `i`'th capture.
+  result = c.matches[i]
+
 when not useUnicode:
   type
     TRune = char
@@ -472,9 +510,17 @@ when not useUnicode:
     inc(i)
   template runeLenAt(s, i: expr): expr = 1
 
-proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
-  ## this implements a simple PEG interpreter. Thanks to superoperators it
-  ## has competitive performance nevertheless.
+  proc isAlpha(a: char): bool {.inline.} = return a in {'a'..'z','A'..'Z'}
+  proc isUpper(a: char): bool {.inline.} = return a in {'A'..'Z'}
+  proc isLower(a: char): bool {.inline.} = return a in {'a'..'z'}
+  proc isTitle(a: char): bool {.inline.} = return false
+  proc isWhiteSpace(a: char): bool {.inline.} = return a in {' ', '\9'..'\13'}
+
+proc rawMatch*(s: string, p: TPeg, start: int, c: var TCaptures): int {.
+               nosideEffect, rtl, extern: "npegs$1".} =
+  ## low-level matching proc that implements the PEG interpreter. Use this 
+  ## for maximum efficiency (every other PEG operation ends up calling this
+  ## proc).
   ## Returns -1 if it does not match, else the length of the match
   case p.kind
   of pkEmpty: result = 0 # match of length 0
@@ -486,6 +532,51 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
       result = runeLenAt(s, start)
     else:
       result = -1
+  of pkLetter: 
+    if s[start] != '\0':
+      var a: TRune
+      result = start
+      fastRuneAt(s, result, a)
+      if isAlpha(a): dec(result, start)
+      else: result = -1
+    else:
+      result = -1
+  of pkLower: 
+    if s[start] != '\0':
+      var a: TRune
+      result = start
+      fastRuneAt(s, result, a)
+      if isLower(a): dec(result, start)
+      else: result = -1
+    else:
+      result = -1
+  of pkUpper: 
+    if s[start] != '\0':
+      var a: TRune
+      result = start
+      fastRuneAt(s, result, a)
+      if isUpper(a): dec(result, start)
+      else: result = -1
+    else:
+      result = -1
+  of pkTitle: 
+    if s[start] != '\0':
+      var a: TRune
+      result = start
+      fastRuneAt(s, result, a)
+      if isTitle(a): dec(result, start) 
+      else: result = -1
+    else:
+      result = -1
+  of pkWhitespace: 
+    if s[start] != '\0':
+      var a: TRune
+      result = start
+      fastRuneAt(s, result, a)
+      if isWhitespace(a): dec(result, start)
+      else: result = -1
+    else:
+      result = -1
   of pkGreedyAny:
     result = len(s) - start
   of pkNewLine:
@@ -537,14 +628,14 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
   of pkNonTerminal:
     var oldMl = c.ml
     when false: echo "enter: ", p.nt.name
-    result = m(s, p.nt.rule, start, c)
+    result = rawMatch(s, p.nt.rule, start, c)
     when false: echo "leave: ", p.nt.name
     if result < 0: c.ml = oldMl
   of pkSequence:
     var oldMl = c.ml  
     result = 0
     for i in 0..high(p.sons):
-      var x = m(s, p.sons[i], start+result, c)
+      var x = rawMatch(s, p.sons[i], start+result, c)
       if x < 0:
         c.ml = oldMl
         result = -1
@@ -553,14 +644,14 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
   of pkOrderedChoice:
     var oldMl = c.ml
     for i in 0..high(p.sons):
-      result = m(s, p.sons[i], start, c)
+      result = rawMatch(s, p.sons[i], start, c)
       if result >= 0: break
       c.ml = oldMl
   of pkSearch:
     var oldMl = c.ml
     result = 0
     while start+result < s.len:
-      var x = m(s, p.sons[0], start+result, c)
+      var x = rawMatch(s, p.sons[0], start+result, c)
       if x >= 0:
         inc(result, x)
         return
@@ -572,7 +663,7 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
     inc(c.ml)
     result = 0
     while start+result < s.len:
-      var x = m(s, p.sons[0], start+result, c)
+      var x = rawMatch(s, p.sons[0], start+result, c)
       if x >= 0:
         if idx < maxSubpatterns:
           c.matches[idx] = (start, start+result-1)
@@ -585,7 +676,7 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
   of pkGreedyRep:
     result = 0
     while true:
-      var x = m(s, p.sons[0], start+result, c)
+      var x = rawMatch(s, p.sons[0], start+result, c)
       # if x == 0, we have an endless loop; so the correct behaviour would be
       # not to break. But endless loops can be easily introduced:
       # ``(comment / \w*)*`` is such an example. Breaking for x == 0 does the
@@ -600,15 +691,15 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
     result = 0
     while contains(p.charChoice^, s[start+result]): inc(result)
   of pkOption:
-    result = max(0, m(s, p.sons[0], start, c))
+    result = max(0, rawMatch(s, p.sons[0], start, c))
   of pkAndPredicate:
     var oldMl = c.ml
-    result = m(s, p.sons[0], start, c)
+    result = rawMatch(s, p.sons[0], start, c)
     if result >= 0: result = 0 # do not consume anything
     else: c.ml = oldMl
   of pkNotPredicate:
     var oldMl = c.ml
-    result = m(s, p.sons[0], start, c)
+    result = rawMatch(s, p.sons[0], start, c)
     if result < 0: result = 0
     else:
       c.ml = oldMl
@@ -616,7 +707,7 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
   of pkCapture:
     var idx = c.ml # reserve a slot for the subpattern
     inc(c.ml)
-    result = m(s, p.sons[0], start, c)
+    result = rawMatch(s, p.sons[0], start, c)
     if result >= 0:
       if idx < maxSubpatterns:
         c.matches[idx] = (start, start+result-1)
@@ -629,7 +720,7 @@ proc m(s: string, p: TPeg, start: int, c: var TMatchClosure): int =
     var n: TPeg
     n.kind = succ(pkTerminal, ord(p.kind)-ord(pkBackRef)) 
     n.term = s.copy(a, b)
-    result = m(s, n, start, c)
+    result = rawMatch(s, n, start, c)
   of pkRule, pkList: assert false
 
 proc match*(s: string, pattern: TPeg, matches: var openarray[string],
@@ -638,8 +729,8 @@ proc match*(s: string, pattern: TPeg, matches: var openarray[string],
   ## the captured substrings in the array ``matches``. If it does not
   ## match, nothing is written into ``matches`` and ``false`` is
   ## returned.
-  var c: TMatchClosure
-  result = m(s, pattern, start, c) == len(s) -start
+  var c: TCaptures
+  result = rawMatch(s, pattern, start, c) == len(s) -start
   if result:
     for i in 0..c.ml-1:
       matches[i] = copy(s, c.matches[i][0], c.matches[i][1])
@@ -647,8 +738,8 @@ proc match*(s: string, pattern: TPeg, matches: var openarray[string],
 proc match*(s: string, pattern: TPeg, 
             start = 0): bool {.nosideEffect, rtl, extern: "npegs$1".} =
   ## returns ``true`` if ``s`` matches the ``pattern`` beginning from ``start``.
-  var c: TMatchClosure
-  result = m(s, pattern, start, c) == len(s)-start
+  var c: TCaptures
+  result = rawMatch(s, pattern, start, c) == len(s)-start
 
 proc matchLen*(s: string, pattern: TPeg, matches: var openarray[string],
                start = 0): int {.nosideEffect, rtl, extern: "npegs$1Capture".} =
@@ -656,8 +747,8 @@ proc matchLen*(s: string, pattern: TPeg, matches: var openarray[string],
   ## if there is no match, -1 is returned. Note that a match length
   ## of zero can happen. It's possible that a suffix of `s` remains
   ## that does not belong to the match.
-  var c: TMatchClosure
-  result = m(s, pattern, start, c)
+  var c: TCaptures
+  result = rawMatch(s, pattern, start, c)
   if result >= 0:
     for i in 0..c.ml-1:
       matches[i] = copy(s, c.matches[i][0], c.matches[i][1])
@@ -668,8 +759,8 @@ proc matchLen*(s: string, pattern: TPeg,
   ## if there is no match, -1 is returned. Note that a match length
   ## of zero can happen. It's possible that a suffix of `s` remains
   ## that does not belong to the match.
-  var c: TMatchClosure
-  result = m(s, pattern, start, c)
+  var c: TCaptures
+  result = rawMatch(s, pattern, start, c)
 
 proc find*(s: string, pattern: TPeg, matches: var openarray[string],
            start = 0): int {.nosideEffect, rtl, extern: "npegs$1Capture".} =
@@ -681,6 +772,18 @@ proc find*(s: string, pattern: TPeg, matches: var openarray[string],
   return -1
   # could also use the pattern here: (!P .)* P
   
+proc findBounds*(s: string, pattern: TPeg, matches: var openarray[string],
+                 start = 0): tuple[first, last: int] {.
+                 nosideEffect, rtl, extern: "npegs$1Capture".} =
+  ## returns the starting position and end position of ``pattern`` in ``s`` 
+  ## and the captured
+  ## substrings in the array ``matches``. If it does not match, nothing
+  ## is written into ``matches`` and (-1,0) is returned.
+  for i in start .. s.len-1:
+    var L = matchLen(s, pattern, matches, i)
+    if L >= 0: return (i, i+L-1)
+  return (-1, 0)
+  
 proc find*(s: string, pattern: TPeg, 
            start = 0): int {.nosideEffect, rtl, extern: "npegs$1".} =
   ## returns the starting position of ``pattern`` in ``s``. If it does not
@@ -1351,6 +1454,11 @@ proc primary(p: var TPegParser): TPeg =
     of "a": result = charset({'a'..'z', 'A'..'Z'})
     of "A": result = charset({'\1'..'\xff'} - {'a'..'z', 'A'..'Z'})
     of "ident": result = pegs.ident
+    of "letter": result = UnicodeLetter()
+    of "upper": result = UnicodeUpper()
+    of "lower": result = UnicodeLower()
+    of "title": result = UnicodeTitle()
+    of "white": result = UnicodeWhitespace()
     else: pegError(p, "unknown built-in: " & p.tok.literal)
     getTok(p)
   of tkEscaped:
@@ -1439,9 +1547,12 @@ proc rawParse(p: var TPegParser): TPeg =
     elif ntUsed notin nt.flags and i > 0:
       pegError(p, "unused rule: " & nt.name, nt.line, nt.col)
 
-proc parsePeg*(input: string, filename = "pattern", line = 1, col = 0): TPeg =
+proc parsePeg*(pattern: string, filename = "pattern", line = 1, col = 0): TPeg =
+  ## constructs a TPeg object from `pattern`. `filename`, `line`, `col` are
+  ## used for error messages, but they only provide start offsets. `parsePeg`
+  ## keeps track of line and column numbers within `pattern`.
   var p: TPegParser
-  init(TPegLexer(p), input, filename, line, col)
+  init(TPegLexer(p), pattern, filename, line, col)
   p.tok.kind = tkInvalid
   p.tok.modifier = modNone
   p.tok.literal = ""
@@ -1505,9 +1616,9 @@ when isMainModule:
   expr.rule = sequence(capture(ident), *sequence(
                 nonterminal(ws), term('+'), nonterminal(ws), nonterminal(expr)))
   
-  var c: TMatchClosure
+  var c: TCaptures
   var s = "a+b +  c +d+e+f"
-  assert m(s, expr.rule, 0, c) == len(s)
+  assert rawMatch(s, expr.rule, 0, c) == len(s)
   var a = ""
   for i in 0..c.ml-1:
     a.add(copy(s, c.matches[i][0], c.matches[i][1]))
@@ -1559,4 +1670,10 @@ when isMainModule:
   else:
     assert false
   
-
+  assert match("eine übersicht und außerdem", peg"(\letter \white*)+")
+  # ß is not a lower cased letter?!
+  assert match("eine übersicht und auerdem", peg"(\lower \white*)+")
+  assert match("EINE ÜBERSICHT UND AUSSERDEM", peg"(\upper \white*)+")
+  assert(not match("456678", peg"(\letter)+"))
+  
+  
diff --git a/lib/pure/sockets.nim b/lib/pure/sockets.nim
index 85628db78e..add41afd6f 100755
--- a/lib/pure/sockets.nim
+++ b/lib/pure/sockets.nim
@@ -9,7 +9,7 @@
 
 ## This module implements a simple portable type-safe sockets layer.
 
-import os
+import os, parseutils
 
 when defined(Windows):
   import winlean
@@ -146,18 +146,66 @@ proc listen*(socket: TSocket, attempts = 5) =
   ## listens to socket.
   if listen(cint(socket), cint(attempts)) < 0'i32: OSError()
 
-proc bindAddr*(socket: TSocket, port = TPort(0)) =
-  ## binds a port number to a socket.
+proc invalidIp4(s: string) {.noreturn, noinline.} =
+  raise newException(EInvalidValue, "invalid ip4 address: " & s)
+
+proc parseIp4*(s: string): int32 = 
+  ## parses an IP version 4 in dotted decimal form like "a.b.c.d".
+  ## Raises EInvalidValue in case of an error.
+  var a, b, c, d: int
+  var i = 0
+  var j = parseInt(s, a, i)
+  if j <= 0: invalidIp4(s)
+  inc(i, j)
+  if s[i] == '.': inc(i)
+  else: invalidIp4(s)
+  j = parseInt(s, b, i)
+  if j <= 0: invalidIp4(s)
+  inc(i, j)
+  if s[i] == '.': inc(i)
+  else: invalidIp4(s)
+  j = parseInt(s, c, i)
+  if j <= 0: invalidIp4(s)
+  inc(i, j)
+  if s[i] == '.': inc(i)
+  else: invalidIp4(s)
+  j = parseInt(s, d, i)
+  if j <= 0: invalidIp4(s)
+  inc(i, j)
+  if s[i] != '\0': invalidIp4(s)
+  result = int32(a shl 24 or b shl 16 or c shl 8 or d)
+
+proc bindAddr*(socket: TSocket, port = TPort(0), address = "") =
+  ## binds an address/port number to a socket.
+  ## Use address string in dotted decimal form like "a.b.c.d"
+  ## or leave "" for any address.
   var name: Tsockaddr_in
   when defined(Windows):
     name.sin_family = int16(ord(AF_INET))
   else:
     name.sin_family = posix.AF_INET
   name.sin_port = sockets.htons(int16(port))
-  name.sin_addr.s_addr = sockets.htonl(INADDR_ANY)
+  if address == "":
+    name.sin_addr.s_addr = sockets.htonl(INADDR_ANY)
+  else:
+    name.sin_addr.s_addr = parseIp4(address)
   if bindSocket(cint(socket), cast[ptr TSockAddr](addr(name)),
                 sizeof(name)) < 0'i32:
     OSError()
+
+when false:
+  proc bindAddr*(socket: TSocket, port = TPort(0)) =
+    ## binds a port number to a socket.
+    var name: Tsockaddr_in
+    when defined(Windows):
+      name.sin_family = int16(ord(AF_INET))
+    else:
+      name.sin_family = posix.AF_INET
+    name.sin_port = sockets.htons(int16(port))
+    name.sin_addr.s_addr = sockets.htonl(INADDR_ANY)
+    if bindSocket(cint(socket), cast[ptr TSockAddr](addr(name)),
+                  sizeof(name)) < 0'i32:
+      OSError()
   
 proc getSockName*(socket: TSocket): TPort = 
   ## returns the socket's associated port number.
@@ -409,6 +457,33 @@ proc send*(socket: TSocket, data: string) =
   ## sends data to a socket.
   if send(socket, cstring(data), data.len) != data.len: OSError()
 
+when defined(Windows):
+  const 
+    SOCKET_ERROR = -1
+    IOCPARM_MASK = 127
+    IOC_IN = int(-2147483648)
+    FIONBIO = int(IOC_IN or ((sizeof(int) and IOCPARM_MASK) shl 16) or 
+                             (102 shl 8) or 126)
+
+  proc ioctlsocket(s: TWinSocket, cmd: clong, 
+                   argptr: ptr clong): cint {.
+                   stdcall, importc:"ioctlsocket", dynlib: "ws2_32.dll".}
+
+proc setBlocking*(s: TSocket, blocking: bool) =
+  ## sets blocking mode on socket
+  when defined(Windows):
+    var mode = clong(ord(not blocking)) # 1 for non-blocking, 0 for blocking
+    if SOCKET_ERROR == ioctlsocket(TWinSocket(s), FIONBIO, addr(mode)):
+      OSError()
+  else: # BSD sockets
+    var x: int = fcntl(cint(s), F_GETFL, 0)
+    if x == -1:
+      OSError()
+    else:
+      var mode = if blocking: x and not O_NONBLOCK else: x or O_NONBLOCK
+      if fcntl(cint(s), F_SETFL, mode) == -1:
+        OSError()
+
 when defined(Windows):
   var wsa: TWSADATA
   if WSAStartup(0x0101'i16, wsa) != 0: OSError()
diff --git a/lib/system/excpt.nim b/lib/system/excpt.nim
index 938583bf72..9beeb659e2 100755
--- a/lib/system/excpt.nim
+++ b/lib/system/excpt.nim
@@ -62,10 +62,7 @@ proc popCurrentException {.compilerRtl, inl.} =
 # some platforms have native support for stack traces:
 const
   nativeStackTrace = (defined(macosx) or defined(linux)) and 
-                     not nimrodStackTrace and false
-
-# `nativeStackTrace` does not work for me --> deactivated for now. Maybe for 
-# the next release version.
+                     not nimrodStackTrace
 
 when nativeStacktrace:
   type
diff --git a/rod/docgen.nim b/rod/docgen.nim
index 9da191d8da..abb33a6f1f 100755
--- a/rod/docgen.nim
+++ b/rod/docgen.nim
@@ -756,6 +756,9 @@ proc renderRstToOut(d: PDoc, n: PRstNode): PRope =
   of rnTitle: d.meta[metaTitle] = renderRstToOut(d, n.sons[0])
   else: InternalError("renderRstToOut")
   
+proc checkForFalse(n: PNode): bool = 
+  result = n.kind == nkIdent and IdentEq(n.ident, "false")
+  
 proc generateDoc(d: PDoc, n: PNode) = 
   if n == nil: return 
   case n.kind
@@ -782,7 +785,8 @@ proc generateDoc(d: PDoc, n: PNode) =
     for i in countup(0, sonsLen(n) - 1): generateDoc(d, n.sons[i])
   of nkWhenStmt: 
     # generate documentation for the first branch only:
-    generateDoc(d, lastSon(n.sons[0]))
+    if not checkForFalse(n.sons[0].sons[0]):
+      generateDoc(d, lastSon(n.sons[0]))
   else: nil
 
 proc genSection(d: PDoc, kind: TSymKind) = 
diff --git a/tests/gc/talloc.nim b/tests/gc/talloc.nim
new file mode 100755
index 0000000000..79a8424151
--- /dev/null
+++ b/tests/gc/talloc.nim
@@ -0,0 +1,637 @@
+#
+#
+#            Nimrod's Runtime Library
+#        (c) Copyright 2010 Andreas Rumpf
+#
+#    See the file "copying.txt", included in this
+#    distribution, for details about the copyright.
+#
+
+# Low level allocator for Nimrod. Has been designed to support the GC.
+# TODO: 
+# - eliminate "used" field
+# - make searching for block O(1)
+
+const
+  debugGC = false # we wish to debug the GC...
+  logGC = false
+  traceGC = false # extensive debugging
+  reallyDealloc = true # for debugging purposes this can be set to false
+  cycleGC = true # (de)activate the cycle GC
+  stressGC = false
+  reallyOsDealloc = true
+  coalescRight = true
+  coalescLeft = true
+  overwriteFree = false
+
+# Page size of the system; in most cases 4096 bytes. For exotic OS or
+# CPU this needs to be changed:
+const
+  PageShift = 12
+  PageSize = 1 shl PageShift
+  PageMask = PageSize-1
+
+  MemAlign = 8 # also minimal allocatable memory block
+
+  BitsPerPage = PageSize div MemAlign
+  UnitsPerPage = BitsPerPage div (sizeof(int)*8)
+    # how many ints do we need to describe a page:
+    # on 32 bit systems this is only 16 (!)
+
+  TrunkShift = 9
+  BitsPerTrunk = 1 shl TrunkShift # needs to be power of 2 and divisible by 64
+  TrunkMask = BitsPerTrunk - 1
+  IntsPerTrunk = BitsPerTrunk div (sizeof(int)*8)
+  IntShift = 5 + ord(sizeof(int) == 8) # 5 or 6, depending on int width
+  IntMask = 1 shl IntShift - 1
+
+proc raiseOutOfMem() {.noreturn.} =
+  quit("out of memory")
+
+# ------------ platform specific chunk allocation code -----------------------
+
+when defined(posix): 
+  const
+    PROT_READ  = 1             # page can be read 
+    PROT_WRITE = 2             # page can be written 
+    MAP_PRIVATE = 2            # Changes are private 
+  
+  when defined(linux) or defined(aix):
+    const MAP_ANONYMOUS = 0x20       # don't use a file
+  elif defined(macosx) or defined(bsd):
+    const MAP_ANONYMOUS = 0x1000
+  elif defined(solaris): 
+    const MAP_ANONYMOUS = 0x100
+  else:
+    {.error: "Port memory manager to your platform".}
+
+  proc mmap(adr: pointer, len: int, prot, flags, fildes: cint,
+            off: int): pointer {.header: "<sys/mman.h>".}
+
+  proc munmap(adr: pointer, len: int) {.header: "<sys/mman.h>".}
+  
+  proc osAllocPages(size: int): pointer {.inline.} = 
+    result = mmap(nil, size, PROT_READ or PROT_WRITE, 
+                           MAP_PRIVATE or MAP_ANONYMOUS, -1, 0)
+    if result == nil or result == cast[pointer](-1):
+      raiseOutOfMem()
+      
+  proc osDeallocPages(p: pointer, size: int) {.inline} =
+    when reallyOsDealloc: munmap(p, size)
+  
+elif defined(windows): 
+  const
+    MEM_RESERVE = 0x2000 
+    MEM_COMMIT = 0x1000
+    MEM_TOP_DOWN = 0x100000
+    PAGE_READWRITE = 0x04
+
+    MEM_DECOMMIT = 0x4000
+    MEM_RELEASE = 0x8000
+
+  proc VirtualAlloc(lpAddress: pointer, dwSize: int, flAllocationType,
+                    flProtect: int32): pointer {.
+                    header: "<windows.h>", stdcall.}
+  
+  proc VirtualFree(lpAddress: pointer, dwSize: int, 
+                   dwFreeType: int32) {.header: "<windows.h>", stdcall.}
+  
+  proc osAllocPages(size: int): pointer {.inline.} = 
+    result = VirtualAlloc(nil, size, MEM_RESERVE or MEM_COMMIT,
+                          PAGE_READWRITE)
+    if result == nil: raiseOutOfMem()
+
+  proc osDeallocPages(p: pointer, size: int) {.inline.} = 
+    # according to Microsoft, 0 is the only correct value here:
+    when reallyOsDealloc: VirtualFree(p, 0, MEM_RELEASE)
+
+else: 
+  {.error: "Port memory manager to your platform".}
+
+# --------------------- end of non-portable code -----------------------------
+
+# We manage *chunks* of memory. Each chunk is a multiple of the page size.
+# Each chunk starts at an address that is divisible by the page size. Chunks
+# that are bigger than ``ChunkOsReturn`` are returned back to the operating
+# system immediately.
+
+const
+  ChunkOsReturn = 256 * PageSize
+  InitialMemoryRequest = ChunkOsReturn div 2 # < ChunkOsReturn!
+  SmallChunkSize = PageSize
+
+type 
+  PTrunk = ptr TTrunk
+  TTrunk {.final.} = object 
+    next: PTrunk         # all nodes are connected with this pointer
+    key: int             # start address at bit 0
+    bits: array[0..IntsPerTrunk-1, int] # a bit vector
+  
+  TTrunkBuckets = array[0..1023, PTrunk]
+  TIntSet {.final.} = object 
+    data: TTrunkBuckets
+  
+type
+  TAlignType = biggestFloat
+  TFreeCell {.final, pure.} = object
+    next: ptr TFreeCell  # next free cell in chunk (overlaid with refcount)
+    zeroField: int       # 0 means cell is not used (overlaid with typ field)
+                         # 1 means cell is manually managed pointer
+
+  PChunk = ptr TBaseChunk
+  PBigChunk = ptr TBigChunk
+  PSmallChunk = ptr TSmallChunk
+  TBaseChunk {.pure.} = object
+    prevSize: int        # size of previous chunk; for coalescing
+    size: int            # if < PageSize it is a small chunk
+    used: bool           # later will be optimized into prevSize...
+  
+  TSmallChunk = object of TBaseChunk
+    next, prev: PSmallChunk  # chunks of the same size
+    freeList: ptr TFreeCell
+    free: int            # how many bytes remain    
+    acc: int             # accumulator for small object allocation
+    data: TAlignType     # start of usable memory
+  
+  TBigChunk = object of TBaseChunk # not necessarily > PageSize!
+    next: PBigChunk      # chunks of the same (or bigger) size
+    prev: PBigChunk
+    align: int
+    data: TAlignType     # start of usable memory
+
+template smallChunkOverhead(): expr = sizeof(TSmallChunk)-sizeof(TAlignType)
+template bigChunkOverhead(): expr = sizeof(TBigChunk)-sizeof(TAlignType)
+
+proc roundup(x, v: int): int {.inline.} = 
+  result = (x + (v-1)) and not (v-1)
+  assert(result >= x)
+  #return ((-x) and (v-1)) +% x
+
+assert(roundup(14, PageSize) == PageSize)
+assert(roundup(15, 8) == 16)
+assert(roundup(65, 8) == 72)
+
+# ------------- chunk table ---------------------------------------------------
+# We use a PtrSet of chunk starts and a table[Page, chunksize] for chunk
+# endings of big chunks. This is needed by the merging operation. The only
+# remaining operation is best-fit for big chunks. Since there is a size-limit
+# for big chunks (because greater than the limit means they are returned back
+# to the OS), a fixed size array can be used. 
+
+type
+  PLLChunk = ptr TLLChunk
+  TLLChunk {.pure.} = object ## *low-level* chunk
+    size: int                # remaining size
+    acc: int                 # accumulator
+    
+  TAllocator {.final, pure.} = object
+    llmem: PLLChunk
+    currMem, maxMem, freeMem: int # memory sizes (allocated from OS)
+    freeSmallChunks: array[0..SmallChunkSize div MemAlign-1, PSmallChunk]
+    freeChunksList: PBigChunk # XXX make this a datastructure with O(1) access
+    chunkStarts: TIntSet
+   
+proc incCurrMem(a: var TAllocator, bytes: int) {.inline.} = 
+  inc(a.currMem, bytes)
+
+proc decCurrMem(a: var TAllocator, bytes: int) {.inline.} =
+  a.maxMem = max(a.maxMem, a.currMem)
+  dec(a.currMem, bytes)
+
+proc getMaxMem(a: var TAllocator): int =
+  # Since we update maxPagesCount only when freeing pages, 
+  # maxPagesCount may not be up to date. Thus we use the
+  # maximum of these both values here:
+  return max(a.currMem, a.maxMem)
+   
+var
+  allocator: TAllocator
+    
+proc llAlloc(a: var TAllocator, size: int): pointer =
+  # *low-level* alloc for the memory managers data structures. Deallocation
+  # is never done.
+  if a.llmem == nil or size > a.llmem.size:
+    var request = roundup(size+sizeof(TLLChunk), PageSize)
+    a.llmem = cast[PLLChunk](osAllocPages(request))
+    incCurrMem(a, request)
+    a.llmem.size = request - sizeof(TLLChunk)
+    a.llmem.acc = sizeof(TLLChunk)
+  result = cast[pointer](cast[TAddress](a.llmem) + a.llmem.acc)
+  dec(a.llmem.size, size)
+  inc(a.llmem.acc, size)
+  zeroMem(result, size)
+  
+proc IntSetGet(t: TIntSet, key: int): PTrunk = 
+  var it = t.data[key and high(t.data)]
+  while it != nil: 
+    if it.key == key: return it
+    it = it.next
+  result = nil
+
+proc IntSetPut(t: var TIntSet, key: int): PTrunk = 
+  result = IntSetGet(t, key)
+  if result == nil:
+    result = cast[PTrunk](llAlloc(allocator, sizeof(result^)))
+    result.next = t.data[key and high(t.data)]
+    t.data[key and high(t.data)] = result
+    result.key = key
+
+proc Contains(s: TIntSet, key: int): bool = 
+  var t = IntSetGet(s, key shr TrunkShift)
+  if t != nil: 
+    var u = key and TrunkMask
+    result = (t.bits[u shr IntShift] and (1 shl (u and IntMask))) != 0
+  else: 
+    result = false
+  
+proc Incl(s: var TIntSet, key: int) = 
+  var t = IntSetPut(s, key shr TrunkShift)
+  var u = key and TrunkMask
+  t.bits[u shr IntShift] = t.bits[u shr IntShift] or (1 shl (u and IntMask))
+
+proc Excl(s: var TIntSet, key: int) = 
+  var t = IntSetGet(s, key shr TrunkShift)
+  if t != nil:
+    var u = key and TrunkMask
+    t.bits[u shr IntShift] = t.bits[u shr IntShift] and not
+        (1 shl (u and IntMask))
+
+proc ContainsOrIncl(s: var TIntSet, key: int): bool = 
+  var t = IntSetGet(s, key shr TrunkShift)
+  if t != nil: 
+    var u = key and TrunkMask
+    result = (t.bits[u shr IntShift] and (1 shl (u and IntMask))) != 0
+    if not result: 
+      t.bits[u shr IntShift] = t.bits[u shr IntShift] or
+          (1 shl (u and IntMask))
+  else: 
+    Incl(s, key)
+    result = false
+   
+# ------------- chunk management ----------------------------------------------
+proc pageIndex(c: PChunk): int {.inline.} = 
+  result = cast[TAddress](c) shr PageShift
+
+proc pageIndex(p: pointer): int {.inline.} = 
+  result = cast[TAddress](p) shr PageShift
+
+proc pageAddr(p: pointer): PChunk {.inline.} = 
+  result = cast[PChunk](cast[TAddress](p) and not PageMask)
+  assert(Contains(allocator.chunkStarts, pageIndex(result)))
+
+var lastSize = PageSize
+
+proc requestOsChunks(a: var TAllocator, size: int): PBigChunk = 
+  incCurrMem(a, size)
+  inc(a.freeMem, size)
+  result = cast[PBigChunk](osAllocPages(size))
+  assert((cast[TAddress](result) and PageMask) == 0)
+  #zeroMem(result, size)
+  result.next = nil
+  result.prev = nil
+  result.used = false
+  result.size = size
+  # update next.prevSize:
+  var nxt = cast[TAddress](result) +% size
+  assert((nxt and PageMask) == 0)
+  var next = cast[PChunk](nxt)
+  if pageIndex(next) in a.chunkStarts:
+    #echo("Next already allocated!")
+    next.prevSize = size
+  # set result.prevSize:
+  var prv = cast[TAddress](result) -% lastSize
+  assert((nxt and PageMask) == 0)
+  var prev = cast[PChunk](prv)
+  if pageIndex(prev) in a.chunkStarts and prev.size == lastSize:
+    #echo("Prev already allocated!")
+    result.prevSize = lastSize
+  else:
+    result.prevSize = 0 # unknown
+  lastSize = size # for next request
+
+proc freeOsChunks(a: var TAllocator, p: pointer, size: int) = 
+  # update next.prevSize:
+  var c = cast[PChunk](p)
+  var nxt = cast[TAddress](p) +% c.size
+  assert((nxt and PageMask) == 0)
+  var next = cast[PChunk](nxt)
+  if pageIndex(next) in a.chunkStarts:
+    next.prevSize = 0 # XXX used
+  excl(a.chunkStarts, pageIndex(p))
+  osDeallocPages(p, size)
+  decCurrMem(a, size)
+  dec(a.freeMem, size)
+  #c_fprintf(c_stdout, "[Alloc] back to OS: %ld\n", size)
+
+proc isAccessible(p: pointer): bool {.inline.} = 
+  result = Contains(allocator.chunkStarts, pageIndex(p))
+
+proc contains[T](list, x: T): bool = 
+  var it = list
+  while it != nil:
+    if it == x: return true
+    it = it.next
+    
+when false:
+  proc writeFreeList(a: TAllocator) =
+    var it = a.freeChunksList
+    c_fprintf(c_stdout, "freeChunksList: %p\n", it)
+    while it != nil: 
+      c_fprintf(c_stdout, "it: %p, next: %p, prev: %p\n", 
+                it, it.next, it.prev)
+      it = it.next
+
+proc ListAdd[T](head: var T, c: T) {.inline.} = 
+  assert(c notin head)
+  assert c.prev == nil
+  assert c.next == nil
+  c.next = head
+  if head != nil: 
+    assert head.prev == nil
+    head.prev = c
+  head = c
+
+proc ListRemove[T](head: var T, c: T) {.inline.} =
+  assert(c in head)
+  if c == head: 
+    head = c.next
+    assert c.prev == nil
+    if head != nil: head.prev = nil
+  else:
+    assert c.prev != nil
+    c.prev.next = c.next
+    if c.next != nil: c.next.prev = c.prev
+  c.next = nil
+  c.prev = nil
+  
+proc isSmallChunk(c: PChunk): bool {.inline.} = 
+  return c.size <= SmallChunkSize-smallChunkOverhead()
+  #return c.size < SmallChunkSize
+  
+proc chunkUnused(c: PChunk): bool {.inline.} = 
+  result = not c.used
+  
+proc updatePrevSize(a: var TAllocator, c: PBigChunk, 
+                    prevSize: int) {.inline.} = 
+  var ri = cast[PChunk](cast[TAddress](c) +% c.size)
+  assert((cast[TAddress](ri) and PageMask) == 0)
+  if isAccessible(ri):
+    ri.prevSize = prevSize
+  
+proc freeBigChunk(a: var TAllocator, c: PBigChunk) = 
+  var c = c
+  assert(c.size >= PageSize)
+  inc(a.freeMem, c.size)
+  when coalescRight:
+    var ri = cast[PChunk](cast[TAddress](c) +% c.size)
+    assert((cast[TAddress](ri) and PageMask) == 0)
+    if isAccessible(ri) and chunkUnused(ri):
+      assert(not isSmallChunk(ri))
+      if not isSmallChunk(ri):
+        ListRemove(a.freeChunksList, cast[PBigChunk](ri))
+        inc(c.size, ri.size)
+        excl(a.chunkStarts, pageIndex(ri))
+  when coalescLeft:
+    if c.prevSize != 0: 
+      var le = cast[PChunk](cast[TAddress](c) -% c.prevSize)
+      assert((cast[TAddress](le) and PageMask) == 0)
+      if isAccessible(le) and chunkUnused(le):
+        assert(not isSmallChunk(le))
+        if not isSmallChunk(le):
+          ListRemove(a.freeChunksList, cast[PBigChunk](le))
+          inc(le.size, c.size)
+          excl(a.chunkStarts, pageIndex(c))
+          c = cast[PBigChunk](le)
+
+  if c.size < ChunkOsReturn: 
+    incl(a.chunkStarts, pageIndex(c))
+    updatePrevSize(a, c, c.size)
+    ListAdd(a.freeChunksList, c)
+    c.used = false
+  else:
+    freeOsChunks(a, c, c.size)
+
+proc splitChunk(a: var TAllocator, c: PBigChunk, size: int) = 
+  var rest = cast[PBigChunk](cast[TAddress](c) +% size)
+  assert(rest notin a.freeChunksList)
+  #  c_fprintf(c_stdout, "to add: %p\n", rest)
+  #  writeFreeList(allocator)
+  #  assert false
+  rest.size = c.size - size
+  rest.used = false
+  rest.next = nil
+  rest.prev = nil
+  rest.prevSize = size
+  updatePrevSize(a, c, rest.size)
+  c.size = size
+  incl(a.chunkStarts, pageIndex(rest))
+  ListAdd(a.freeChunksList, rest)
+
+proc getBigChunk(a: var TAllocator, size: int): PBigChunk = 
+  # use first fit for now:
+  assert((size and PageMask) == 0)
+  assert(size > 0)
+  result = a.freeChunksList
+  block search:
+    while result != nil:
+      #if not chunkUnused(result): 
+      #  c_fprintf(c_stdout, "%lld\n", int(result.used))
+      assert chunkUnused(result)
+      if result.size == size: 
+        ListRemove(a.freeChunksList, result)
+        break search
+      elif result.size > size:
+        #c_fprintf(c_stdout, "res size: %lld; size: %lld\n", result.size, size)
+        ListRemove(a.freeChunksList, result)
+        splitChunk(a, result, size)
+        break search
+      result = result.next
+      assert result != a.freeChunksList
+    if size < InitialMemoryRequest: 
+      result = requestOsChunks(a, InitialMemoryRequest)
+      splitChunk(a, result, size)
+    else:
+      result = requestOsChunks(a, size)
+  result.prevSize = 0 # XXX why is this needed?
+  result.used = true
+  incl(a.chunkStarts, pageIndex(result))
+  dec(a.freeMem, size)
+
+proc getSmallChunk(a: var TAllocator): PSmallChunk = 
+  var res = getBigChunk(a, PageSize)
+  assert res.prev == nil
+  assert res.next == nil
+  result = cast[PSmallChunk](res)
+
+# -----------------------------------------------------------------------------
+
+proc getCellSize(p: pointer): int {.inline.} = 
+  var c = pageAddr(p)
+  result = c.size
+  
+proc rawAlloc(a: var TAllocator, requestedSize: int): pointer =
+  assert(roundup(65, 8) == 72)
+  assert requestedSize >= sizeof(TFreeCell)
+  var size = roundup(requestedSize, MemAlign)
+  #c_fprintf(c_stdout, "alloc; size: %ld; %ld\n", requestedSize, size)
+  if size <= SmallChunkSize-smallChunkOverhead(): 
+    # allocate a small block: for small chunks, we use only its next pointer
+    var s = size div MemAlign
+    var c = a.freeSmallChunks[s]
+    if c == nil: 
+      c = getSmallChunk(a)
+      c.freeList = nil
+      assert c.size == PageSize
+      c.size = size
+      c.acc = size
+      c.free = SmallChunkSize - smallChunkOverhead() - size
+      c.next = nil
+      c.prev = nil
+      ListAdd(a.freeSmallChunks[s], c)
+      result = addr(c.data)
+      assert((cast[TAddress](result) and (MemAlign-1)) == 0)
+    else:
+      assert c.next != c
+      #if c.size != size:
+      #  c_fprintf(c_stdout, "csize: %lld; size %lld\n", c.size, size)
+      assert c.size == size
+      if c.freeList == nil:
+        assert(c.acc + smallChunkOverhead() + size <= SmallChunkSize) 
+        result = cast[pointer](cast[TAddress](addr(c.data)) +% c.acc)
+        inc(c.acc, size)      
+      else:
+        result = c.freeList
+        assert(c.freeList.zeroField == 0)
+        c.freeList = c.freeList.next
+      dec(c.free, size)
+      assert((cast[TAddress](result) and (MemAlign-1)) == 0)
+    if c.free < size: 
+      ListRemove(a.freeSmallChunks[s], c)
+  else:
+    size = roundup(requestedSize+bigChunkOverhead(), PageSize)
+    # allocate a large block
+    var c = getBigChunk(a, size)
+    assert c.prev == nil
+    assert c.next == nil
+    assert c.size == size
+    result = addr(c.data)
+    assert((cast[TAddress](result) and (MemAlign-1)) == 0)
+  assert(isAccessible(result))
+
+proc rawDealloc(a: var TAllocator, p: pointer) = 
+  var c = pageAddr(p)
+  if isSmallChunk(c):
+    # `p` is within a small chunk:
+    var c = cast[PSmallChunk](c)
+    var s = c.size
+    var f = cast[ptr TFreeCell](p)
+    #echo("setting to nil: ", $cast[TAddress](addr(f.zeroField)))
+    assert(f.zeroField != 0)
+    f.zeroField = 0
+    f.next = c.freeList
+    c.freeList = f
+    when overwriteFree: 
+      # set to 0xff to check for usage after free bugs:
+      c_memset(cast[pointer](cast[int](p) +% sizeof(TFreeCell)), -1'i32, 
+               s -% sizeof(TFreeCell))
+    # check if it is not in the freeSmallChunks[s] list:
+    if c.free < s:
+      assert c notin a.freeSmallChunks[s div memAlign]
+      # add it to the freeSmallChunks[s] array:
+      ListAdd(a.freeSmallChunks[s div memAlign], c)
+      inc(c.free, s)
+    else:
+      inc(c.free, s)
+      if c.free == SmallChunkSize-smallChunkOverhead():
+        ListRemove(a.freeSmallChunks[s div memAlign], c)
+        c.size = SmallChunkSize
+        freeBigChunk(a, cast[PBigChunk](c))
+  else:
+    # set to 0xff to check for usage after free bugs:
+    when overwriteFree: c_memset(p, -1'i32, c.size -% bigChunkOverhead())
+    # free big chunk
+    freeBigChunk(a, cast[PBigChunk](c))
+
+proc isAllocatedPtr(a: TAllocator, p: pointer): bool = 
+  if isAccessible(p):
+    var c = pageAddr(p)
+    if not chunkUnused(c):
+      if isSmallChunk(c):
+        var c = cast[PSmallChunk](c)
+        var offset = (cast[TAddress](p) and PageMask) -% smallChunkOverhead()
+        result = (c.acc >% offset) and (offset %% c.size == 0) and
+          (cast[ptr TFreeCell](p).zeroField >% 1)
+      else:
+        var c = cast[PBigChunk](c)
+        result = p == addr(c.data) and cast[ptr TFreeCell](p).zeroField >% 1
+
+# ---------------------- interface to programs -------------------------------
+
+when true:
+  proc alloc(size: int): pointer =
+    result = rawAlloc(allocator, size+sizeof(TFreeCell))
+    cast[ptr TFreeCell](result).zeroField = 2 # mark it as used
+    #assert(not isAllocatedPtr(allocator, result))
+    result = cast[pointer](cast[TAddress](result) +% sizeof(TFreeCell))
+
+  proc alloc0(size: int): pointer =
+    result = talloc.alloc(size)
+    zeroMem(result, size)
+
+  proc dealloc(p: pointer) =
+    var x = cast[pointer](cast[TAddress](p) -% sizeof(TFreeCell))
+    assert(cast[ptr TFreeCell](x).zeroField == 2)
+    rawDealloc(allocator, x)
+    assert(not isAllocatedPtr(allocator, x))
+
+  proc isAllocatedPtr(p: pointer): bool = 
+    var x = cast[pointer](cast[TAddress](p) -% sizeof(TFreeCell))
+    result = isAllocatedPtr(allocator, x)
+
+  proc ptrSize(p: pointer): int =
+    var x = cast[pointer](cast[TAddress](p) -% sizeof(TFreeCell))
+    result = pageAddr(x).size - sizeof(TFreeCell)
+
+  proc realloc(p: pointer, newsize: int): pointer =
+    if newsize > 0:
+      result = talloc.alloc(newsize)
+      if p != nil:
+        copyMem(result, p, ptrSize(p))
+        talloc.dealloc(p)
+    elif p != nil:
+      talloc.dealloc(p)
+
+  proc countFreeMem(): int =
+    # only used for assertions
+    var it = allocator.freeChunksList
+    while it != nil:
+      inc(result, it.size)
+      it = it.next
+
+  proc getFreeMem(): int = 
+    result = allocator.freeMem
+    #assert(result == countFreeMem())
+
+  proc getTotalMem(): int = return allocator.currMem
+  proc getOccupiedMem(): int = return talloc.getTotalMem() - talloc.getFreeMem()
+
+when isMainModule:
+  const iterations = 4000_000
+  incl(allocator.chunkStarts, 11)
+  assert 11 in allocator.chunkStarts
+  excl(allocator.chunkStarts, 11)
+  assert 11 notin allocator.chunkStarts
+  var p: array [1..iterations, pointer]
+  for i in 4..70:
+    var x = i * 8
+    for j in 1.. iterations:
+      p[j] = talloc.Alloc(x)
+    for j in 1..iterations:
+      assert isAllocatedPtr(p[j])
+    echo($i, " used memory: ", $(allocator.currMem))
+    for j in countdown(iterations, 1):
+      #echo("j: ", $j)
+      talloc.dealloc(p[j])
+      assert(not isAllocatedPtr(allocator, p[j]))
+    echo($i, " after freeing: ", $(allocator.currMem))
+    
diff --git a/tinyc/config.h b/tinyc/config.h
index edaf335bb2..161a3e1783 100755
--- a/tinyc/config.h
+++ b/tinyc/config.h
@@ -7,11 +7,13 @@
 #  define TCC_TARGET_I386
 #  define CONFIG_TCCDIR "."
 #elif defined(__i386__)
+#  define CONFIG_USE_LIBGCC
 #  define TCC_TARGET_I386
 #  define CONFIG_TCCDIR "/usr/local/lib/tcc"
 #  define GCC_MAJOR 4
 #  define HOST_I386 1
 #else
+#  define CONFIG_USE_LIBGCC
 #  define TCC_TARGET_X86_64
 #  define CONFIG_TCCDIR "/usr/local/lib/tcc"
 #  define GCC_MAJOR 4
diff --git a/tools/niminst.nim b/tools/niminst.nim
index 77cdfa16bf..e696b6b93c 100755
--- a/tools/niminst.nim
+++ b/tools/niminst.nim
@@ -84,6 +84,7 @@ proc initConfigData(c: var TConfigData) =
   c.vars = newStringTable(modeStyleInsensitive)
 
 proc skipRoot(f: string): string = 
+  # "abc/def/xyz" --> "def/xyz"
   var i = 0
   result = ""
   for component in split(f, {dirsep, altsep}): 
diff --git a/tools/nimrepl.nim b/tools/nimrepl.nim
index 432ca1356e..220307dba6 100755
--- a/tools/nimrepl.nim
+++ b/tools/nimrepl.nim
@@ -39,7 +39,7 @@ proc destroy(widget: PWidget, data: pgpointer){.cdecl.} =
 proc FileOpenClicked(menuitem: PMenuItem, userdata: pgpointer) {.cdecl.} =
   var path = ChooseFileToOpen(w)
   if path != "":
-    var file: string = readFile(path)
+    var file = readFile(path)
     if file != nil:
       set_text(InputTextBuffer, file, len(file))
     else:
diff --git a/web/news.txt b/web/news.txt
index 4e6c51c56d..25e34dd59f 100755
--- a/web/news.txt
+++ b/web/news.txt
@@ -5,6 +5,8 @@ News
 2010-XX-XX Version 0.8.12 released
 ==================================
 
+Version 0.8.12 has been released! Get it `here <download.html>`_. 
+
 Bugfixes
 --------
 
@@ -21,7 +23,9 @@ Additions
 
 - Added ``re.findAll``, ``pegs.findAll``.
 - Added ``os.findExe``.
-- The Pegs module supports a *captured search loop operator* ``{@}``.
+- Pegs support a *captured search loop operator* ``{@}``.
+- Pegs support new built-ins: ``\letter``, ``\upper``, ``\lower``,
+  ``\title``, ``\white``.
 
 
 2010-10-20 Version 0.8.10 released
@@ -95,9 +99,6 @@ Additions
 2010-03-14 Version 0.8.8 released
 =================================
 
-Version 0.8.8 has been released! Get it `here <download.html>`_. 
-
-
 Bugfixes
 --------
 - The Posix version of ``os.copyFile`` has better error handling.