07:52 <amazigh> it is hidden because it is not public, but the implementation is available on pypi / github under the name: immutables

08:08 <fijal> marmoute: ok, how do I check what is going on? I tried pip install hg to get the newest version

08:16 <fijal> ok, so buffers (continuation of a private chat with cfbolz)

08:16 <fijal> right now we have a) buffer interface stuff (like bytearray/memoryview) b) ffi buffers (of more than one kind too)

08:16 <fijal> there are some competing requirements:

08:17 <fijal> 1. getting address that stays stable. Right now it's done with copying to old generation, which is bad, creates long living list of things with __del__ and old generation immediately

08:17 <cfbolz> fijal: do you mean b) cffi buffers?

08:17 <fijal> 2. any of those operations causes JIT to not look into the functions, because it does not know the types of things

08:17 <fijal> so Bytearray.__init__ is a jit-dont-look operations

08:17 <fijal> cfbolz: yes

08:27 <cfbolz> fijal: it might be interesting to also think about common use cases too

08:28 <fijal> it's a bit hard to say what's "common"

08:28 <fijal> one I'm fighting right now is you have a buffer that you pass to recv_into or send or something like that

08:30 <cfbolz> fijal: ok, that's one ;-)

08:46 <fijal> hard for me to think about others

08:46 <fijal> I'm sure there are

09:03 <fijal> eh, our interp_lsprof is really broken

09:03 <fijal> well, maybe not entirely, but passing a custom timer does not work at all I think

09:05 <fijal> cfbolz: do we have any idea why it tries to do int_w on a custom timer and not float_w?

09:14 <cfbolz> fijal: yes. that's an incompatibility that we didn't fix yet

09:25 <fijal> right