PySyck handles non-alphanumeric strings as references, others not

[thomas@…]

>>> 'a' is 'a'
True
>>> 'a.' is 'a.'
True
>>> syck.dump(['a', 'a'])
'--- \n- a\n- a\n'
>>> syck.dump(['a.', 'a.'])
'--- \n- &id001 a.\n- *id001\n'
>>>

the first two commands show that python itself doesn't distinguish. This is especially problematic since the second case produces a segmentation fault when using syck-0.55-r3 (and surely other older versions).

[xi]

It is possible in PySyck to forbid generating aliases for some specific kinds of identical objects. By default, the following heuristics is used: if an object is a number, or an empty tuple, or an alphanumerical string, aliases are not generated. You can override this by overriding the method Dumper.allow_aliases():

import syck

class MyDumper(syck.Dumper):

    def allow_aliases(self, object):
        if type(object) == str:
            return False
        return syck.Dumper.allow_aliases(self, object)

print syck.dump(['a', 'a'])
print syck.dump(['a.', 'a.'])
print syck.dump(['a', 'a'], Dumper=MyDumper)
print syck.dump(['a.', 'a.'], Dumper=MyDumper)

This code produces:

- a
- a

---
- &id001 a.
- *id001

---
- a
- a

---
- a.
- a.