Bad compare in DbRow class
In python/lib/objects.py, yesterday I replaced the DbRow class' __cmp__() method with __eq__() and __ne__() for campatibility with Python3. However, in Python3, I started getting test failures like: ERROR: test_0110 (test_jdb_freq.Test_jdb_copy_freqs) ---------------------------------------------------------------------- Traceback (most recent call last): File "/home/stuart/devel/jdb/jb3/python/tests/test_jdb_freq.py", line 57, in test_0110 jdb.copy_freqs (p, e) File "/home/stuart/devel/jdb/jb3/python/lib/jdb.py", line 889, in copy_freqs dupl = _copy_freqs (old_entr, new_entr) File "/home/stuart/devel/jdb/jb3/python/lib/jdb.py", line 899, in _copy_freqs finv = freq_inv (old_entr) File "/home/stuart/devel/jdb/jb3/python/lib/jdb.py", line 1012, in freq_inv assert f not in b TypeError: unhashable type: 'Freq' Problem is that jdb.freq_inv uses Freq objects as keys in a dict. Turns out that in Python3, DbRow objects loose the default __hash__() method but in Python2 that doesn't happen. Adding a def __hash__(self): return id(self) to DbRow fixes the problem in Python3. For compabilty I made same change in the Python2 branch although it id not neccessary there since in Py2 __hash__() is inherited from the parent class. However.... Reading the docs for __hash__(), __eq__(), and __cmp__() indicates that having __hash__() defined is wrong, and was wrong in Python2 branch (even befor the __cmp__ replacement) as well: objects that compare "==" must have equal hash values. Possible ways to fix. 1) Ignore. Is a problem if object is used as key in dict *and* the object subsequently modified in a way that changes it's "==" test. We seldom use DbRow objects as dict keys and when we do, I don't think they get changed. (But would need to audit the code to confirm.) 2) Make DbRow objects base compares on only the predefined attributes and make those attributes immutable once set initially. 3) Make two versions of DbRow. One, used for Entr and other objects that have lists and are othewise mutated has no __hash__() and can't be used as dict key. Other version used for Freq et.al. is immutable after creation.
[Imported from JMdictDB Issues Tracker: IS-215/msg534]
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information