Gary Poster <gary.poster@gmail.com> schrieb:
OK. I'll give it a whirl sometime over the next couple of weeks, if that's soon enough for you.
Sure.
FWIW, I'd be strongly tempted to release *without* the generation code, and leave it up to users to switch as they desire.
Fine with me. I'm pretty confident of the other changes I've made, BTW, and there's not exactly a lot of them either.
1) That's particularly pertinent for library bits like this because a tree walker would have to walk over *all* attributes and __getitem__s in order to find instances of things like a blist, which will generally be hidden deep in application objects; or would have to use an iteration protocol like the one that FileStorage provides.
Right. Looking closer at the utility function I've used so far for walking the objects, I think it doesn't nearly cover all the places a dict could hide.
2) Old instances (with lists, not blists) will still work fine with the new code, and in fact should continue to work even with new apis as long as the ordered dict apis use the list apis (like slices) to manipulate the order.
They do.
3) How many people are really using the blist right now anyway in production?
No idea...
Generation code is hard to test in the abstract.
Do we actually have any best practices for that? Viele Grüße, Thomas -- Thomas Lotze · tl@gocept.com gocept gmbh & co. kg · forsterstraße 29 · 06112 halle (saale) · germany http://gocept.com · tel +49 345 1229889 0 · fax +49 345 1229889 1 Zope and Plone consulting and development