Julien Muchembled

> I'm still wondering if there's a way for `WSGIPublisher.publish_module` to return a generator that would yield from what is "pushed" by `response.write` and then what the published method returns...

We are talking about the fact that your case is already broken: with WSGI, it makes no difference between: - using `Response.write` - make your `OFS.Image.File` code instanciate a `io.BytesIO()`...

It looks like the future is ASGI and if one solution for this current issue brings us closer to ASGI, we should choose it. I looked at the WSGI compatibility...

I've just looked at the code of waitress (that's what we chose for WSGI) and it does the same kind of transparent buffering as ZServer. What's strange is that it...

> Otherwise, you use an large amount of ZODB connections (limited by the availability of file descriptors) I was wrong in a previous comment. A ZODB Connection (or a thread)...

Oops, I forgot RelStorage. > bound to one of very few worker threads Because of the GIL and the connection cache, one may indeed limit the number of threads within...

BTW, @d-maurer which WSGI implementations do you use ?

In NEO, we are about to implement partial pack, and even different ways of packing partially. This PR seems to be a deadend because lastPack won't be enough. IStorage is...

> ```python > The bug could be fixed if we change IStorage.loadBefore(oid) to return > > (None, serial) # serial = revision where object was deleted > > instead of...

I understand well the need to distinguish between the 2 cases. But I wondered at least whether I'd have something to fix in NEO, which surprised me because that's something...