On Tue, Jul 27, 2010 at 12:50 PM, Stefan Behnel <[email protected]> wrote: > Carl Witty, 27.07.2010 21:40: >> On Tue, Jul 27, 2010 at 3:55 AM, Stefan Behnel wrote: >>> Given that this is a change that may potentially break stuff and we don't >>> even have tests for it, I think we should not let this delay the release of >>> 0.13 any further than we already did anyway. Most people don't test >>> development releases, so I expect a lot of bug reports after the release of >>> 0.13 that will make 0.13.1 (and maybe even 0.13.2) a close follow-up anyway. >>> >>> However, if someone writes suitable tests for all affected special methods >>> in a timely fashion, I'd be +0 on inclusion, just for the same reason of >>> getting a 0.13.1 out soon anyway, in which we may still rip it back out. >> >> What would you consider suitable tests? All I can think of is to >> write Cython classes that define all the special methods, then write >> Python code that extracts each special method as a bound method and >> then calls it. Can you think of more tests? >> >> I'll try to write these tests tonight. I'm sorry I didn't include >> tests in the patch; it was getting late and I wanted to post what I >> had before I went to bed. > > That's fine, we're just too late in the release cycle (IMHO) to include > potentially unsafe patches, especially when they lack dedicated tests. > > >> What about the two known behavioral changes (__getattr__ turns into >> __getattribute__ (mentioned on the ticket), and docstrings are lost >> (as Robert suspected and I just verified)). Are these enough that you >> don't think the patch should get in to 0.13? > > I'd say so. It breaks Python compatibility only for the sake of performance.
On that note, cdef classes are already non-Python compatible in several ways, so I don't think this is as issue as potentially breaking pure Python code (though of course I'd like them to be more like Python classes). If we ever auto-cdef classes, we'll have to make sure to preserve behavior like this. - Robert _______________________________________________ Cython-dev mailing list [email protected] http://codespeak.net/mailman/listinfo/cython-dev
