Strip invalid byte sequences from input

darcs changes --xml might contain invalid UTF-8 byte sequences,
which breaks XML parsing [http://bugs.darcs.net/issue64].

darcs-to-git

@@ -14,6 +14,7 @@ require 'rexml/document'
 require 'optparse'
 require 'yaml'
 require 'pathname'
+require 'iconv'
 
 # Explicitly setting a time zone would cause darcs to only output in
 # that timezone hence we couldn't get the actual patch TZ
@@ -112,11 +113,16 @@ def run(*args)
   system(*args) || raise("Failed to run: #{args.inspect}")
 end
 
+# cf. Paul Battley, http://po-ru.com/diary/fixing-invalid-utf-8-in-ruby-revisited/
+def validate_utf8(s)
+  return Iconv.iconv('UTF-8//IGNORE', 'UTF-8', (s + ' ') ).first[0..-2]
+end
+
 def output_of(*args)
   puts "Running: #{args.inspect}"
   output = IO.popen(args.map {|a| "'#{a}'"}.join(' '), 'r') { |p| p.read }
   if $?.exitstatus == 0
-    return output
+    return validate_utf8(output)
   else
     raise "Failed to run: #{args.inspect}"
   end