Message17468
Logged In: YES
user_id=699265
It is clear that something which used to work no longer
works, and I'm just wondering if the breakage is necessary.
I'm not sure what all was involved with respect to timeouts
and why the makefile() implementation had to change. I
guess I see 3 problems:
1) Previously functional code breaks. While marshal is
not "general purpose" in the sense it can't deal with
instances, etc. it is still very useful and for data which
doesn't contain many redundant references (where cPickle
reduces them) it is the most efficient way to serialize and
unserialize "native" python types.
2) The performance of everything which uses the results of
socket.makefile() now suffers the cost of a pure-python
implementation where it had perviously been optimized as a C
implementation on platforms which supported it (most).
3) Any other 3rd party C extensions which use the
PyFile_Check (and would previously have worked) break.
Is there another way to solve the timeout problem? Breaking
something which has worked since 1.5 somehow feels wrong
to me. |
|
Date |
User |
Action |
Args |
2007-08-23 14:15:31 | admin | link | issue780354 messages |
2007-08-23 14:15:31 | admin | create | |
|