hgkw/keyword.py
branchstable
changeset 855 eca16536349a
parent 854 3450d7f7d1a3
child 864 2cadbd48a31b
--- a/hgkw/keyword.py	Mon Nov 01 10:24:07 2010 +0100
+++ b/hgkw/keyword.py	Fri Nov 05 10:01:23 2010 +0000
@@ -170,14 +170,25 @@
                                   for k, v in kwmaps)
         else:
             self.templates = _defaultkwmaps(self.ui)
-        escaped = '|'.join(map(re.escape, self.templates.keys()))
-        self.re_kw = re.compile(r'\$(%s)\$' % escaped)
-        self.re_kwexp = re.compile(r'\$(%s): [^$\n\r]*? \$' % escaped)
-
         templatefilters.filters.update({'utcdate': utcdate,
                                         'svnisodate': svnisodate,
                                         'svnutcdate': svnutcdate})
 
+    @util.propertycache
+    def escape(self):
+        '''Returns bar-separated and escaped keywords.'''
+        return '|'.join(map(re.escape, self.templates.keys()))
+
+    @util.propertycache
+    def rekw(self):
+        '''Returns regex for unexpanded keywords.'''
+        return re.compile(r'\$(%s)\$' % self.escape)
+
+    @util.propertycache
+    def rekwexp(self):
+        '''Returns regex for expanded keywords.'''
+        return re.compile(r'\$(%s): [^$\n\r]*? \$' % self.escape)
+
     def substitute(self, data, path, ctx, subfunc):
         '''Replaces keywords in data with expanded template.'''
         def kwsub(mobj):
@@ -191,11 +202,15 @@
             return '$%s: %s $' % (kw, ekw)
         return subfunc(kwsub, data)
 
+    def linkctx(self, path, fileid):
+        '''Similar to filelog.linkrev, but returns a changectx.'''
+        return self.repo.filectx(path, fileid=fileid).changectx()
+
     def expand(self, path, node, data):
         '''Returns data with keywords expanded.'''
         if not self.restrict and self.match(path) and not util.binary(data):
-            ctx = self.repo.filectx(path, fileid=node).changectx()
-            return self.substitute(data, path, ctx, self.re_kw.sub)
+            ctx = self.linkctx(path, node)
+            return self.substitute(data, path, ctx, self.rekw.sub)
         return data
 
     def iskwfile(self, cand, ctx):
@@ -212,8 +227,8 @@
         kwcmd = self.restrict and lookup # kwexpand/kwshrink
         if self.restrict or expand and lookup:
             mf = ctx.manifest()
-        fctx = ctx
-        subn = (self.restrict or rekw) and self.re_kw.subn or self.re_kwexp.subn
+        lctx = ctx
+        re_kw = (self.restrict or rekw) and self.rekw or self.rekwexp
         msg = (expand and _('overwriting %s expanding keywords\n')
                or _('overwriting %s shrinking keywords\n'))
         for f in candidates:
@@ -225,12 +240,12 @@
                 continue
             if expand:
                 if lookup:
-                    fctx = self.repo.filectx(f, fileid=mf[f]).changectx()
-                data, found = self.substitute(data, f, fctx, subn)
+                    lctx = self.linkctx(f, mf[f])
+                data, found = self.substitute(data, f, lctx, re_kw.subn)
             elif self.restrict:
-                found = self.re_kw.search(data)
+                found = re_kw.search(data)
             else:
-                data, found = _shrinktext(data, subn)
+                data, found = _shrinktext(data, re_kw.subn)
             if found:
                 self.ui.note(msg % f)
                 self.repo.wwrite(f, data, ctx.flags(f))
@@ -242,7 +257,7 @@
     def shrink(self, fname, text):
         '''Returns text with all keyword substitutions removed.'''
         if self.match(fname) and not util.binary(text):
-            return _shrinktext(text, self.re_kwexp.sub)
+            return _shrinktext(text, self.rekwexp.sub)
         return text
 
     def shrinklines(self, fname, lines):
@@ -250,7 +265,7 @@
         if self.match(fname):
             text = ''.join(lines)
             if not util.binary(text):
-                return _shrinktext(text, self.re_kwexp.sub).splitlines(True)
+                return _shrinktext(text, self.rekwexp.sub).splitlines(True)
         return lines
 
     def wread(self, fname, data):