Hi and happy new years !
2019 starts quite well: I'm done with the refactoring :-) Everything has
been pushed this morning ( I wish I would have been able to do it
yesterday, but at 2AM, after a party I was attending was over, I was too
tired to figure oiut what was wrong, so I just added a comment in the
code saying 'This is here that you need to look at' )
So, it's a big refactoring, and it's not completely over. As a sum up,
here are the big changes:
- we do encode operations in revert order. That means we don't
pre-compute the PDU size before feeding the buffer, we feed a
fixed-sized, expandable buffer from the bottom. The idea is to avoid
computing a value, and to use a buffer that will be stored in the
ThreadLocalStorage later one, avoiding an allocation for each operation
(less GC, less CPU)
- The decorators are GONE ! We use factories instead, which make it
easier to add a new control or extended operation. Those factories are
used to create a new Control/ExtOp/IntermediateResp, encode and decode
their values. The code is smaller (around 8000 lines of code have been
removed, not counting tests), and hopefully better.
- Many bugs were fixed in controls/extop. It's quite surprising to find
out that some of the controls were silply not properly encoded or decoded !
- The server wasn't properly managing exceptions, and the
NoticeOfDisconnect was never received by the client, simply because the
server was closing the connection *before* this extendedResponse was
sent. That made the server-integ last longer than expected, most of the
tests that expected such a disconnection were waiting until the default
timeout (5 seconds most of the time). As a result, tests are 2mins faster.
There are things I need to complete though (even if the API is usable as is)
- a few renaming has to be done (nothing biggy, and it has no impact on
the users)
- ThreadLocalStorage has to be used to store the encoder buffer
- Ideally speaking, I'd like to add a post-decoding action. It's a big
technical, but the thing is that for some operations, we can't really
differ the creation of an message/Control/ExtOp instance, because if we
do, we may not be able to have this instance created at all. This is
typically the case for Control/ExtOp without a name - yes, response may
not have an OID - and without a value. So we create what I called an
OpaqueControl/OpaqueExtendedResponse, then if we have a name, we create
the proper control/extOp instance. This is a kind of waste. With a
post-decoding action, we will be able to only create the proper instance
once and for all
- I want to make the Asn1Decoder a statless class
I don't think those modifications are big, and more important, I can do
them incrementally. I expect to be done by the end of this week.
The JNDI code is still present, and will be removed later (maybe in a
next release).
So this is it, I think we will be able to cut a release this week-end or
earlier, in the mean time, feel free to test the current code !
Thanks!
