# HG changeset patch # User Steve Losh # Date 1251322207 14400 # Node ID 906a8111a48ad9ac66f44af316bdddae17a5686c # Parent 0b4ae60b9015317da6f245a2bc8c1341658ad451 Formatting and whitespace. diff -r 0b4ae60b9015 -r 906a8111a48a t.py --- a/t.py Wed Aug 26 17:28:25 2009 -0400 +++ b/t.py Wed Aug 26 17:30:07 2009 -0400 @@ -15,13 +15,13 @@ def _task_from_taskline(taskline): """Parse a taskline (from a task file) and return a task. - + A taskline should be in the format: - + summary text ... | meta1:meta1_value,meta2:meta2_value,... - + The task returned will be a dictionary such as: - + { 'id': , 'text': , ... other metadata ... } @@ -36,10 +36,10 @@ def _prefixes(ids): """return a mapping of ids to prefixes - + Each prefix will be the shortest possible substring of the ID that can uniquely identify it among the given group of IDs. - + """ prefixes = {} @@ -71,7 +71,7 @@ def add_task(self, text): id = _hash(text) self.tasks[id] = {'id': id, 'text': text} - + def write(self): filemap = (('tasks', self.name), ('done', '.%s.done' % self.name)) for kind, filename in filemap: @@ -102,10 +102,10 @@ self.done[task['id']] = task else: raise AmbiguousPrefix - + def delete_finished(self): self.done = {} - + def build_parser(): parser = OptionParser() @@ -128,20 +128,21 @@ help="print more detailed output (full task ids, etc)") return parser -(options, args) = build_parser().parse_args() - -td = TaskDict(options.name) -text = ' '.join(args) +if __name__ == '__main__': + (options, args) = build_parser().parse_args() + + td = TaskDict(options.name) + text = ' '.join(args) + + if options.finish: + td.finish_task(options.finish) + td.write() + elif options.delete_finished: + td.delete_finished() + td.write() + elif text: + td.add_task(text) + td.write() + else: + td.print_list(verbose=options.verbose) -if options.finish: - td.finish_task(options.finish) - td.write() -elif options.delete_finished: - td.delete_finished() - td.write() -elif text: - td.add_task(text) - td.write() -else: - td.print_list(verbose=options.verbose) -