< Previous by Date Date Index Next by Date >
< Previous in Thread Thread Index  

Re: [reSIProcate] 1.0 release candidate


RFC4475TortureTests is nowhere near complete. The skeleton is there, but there is very little meat. I fully expect this thing to be > 10K lines of code before it is done. (Just look at the first three test cases and extrapolate) This will take some time to complete (I am not even comfortable with slating it for the 1.1 release, seeing as how we have bigger fish to fry; ie outbound and gruu.)

I would love to see issue 13 addressed too, but I don't know if we have the time to choose a fix, and test that fix completely. What has been thrown around so far sounds risky; instead of a little bit of processor spin, the connections could starve. If someone can step up and say "I know how to fix this, I know it will work, and I can write test-cases to prove it.", that may change my mind on the matter. But otherwise, we probably need to take a more conservative approach.

Best regards,
Byron Campen

Great stuff.

I'd love to see at least 13 addressed in this release.  : )

FYI - testTimer and testAppTimer periodically fail in windows for some
reason (some strange timing issue) - but I don't think this is anything
new.

RFC4475TortureTests run without any asserts - does this means it passes? Are there any define's required in order to get all RFC4475TortureTests
to pass?

All other tests in rutil/test and stack/test pass.

Scott

-----Original Message-----
From: resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx
[mailto:resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of
Byron Campen
Sent: Friday, August 18, 2006 1:12 PM
To: resiprocate-devel
Subject: [reSIProcate] 1.0 release candidate


Most of the open issues that we have been discussing have been
addressed at this point, but we have still not fixed/come to a
decision
on the following:


4.       UDP Transport enhancement patch to peek message size and
allocate buffer appropriately.
4.       Not fixed (the MSG_PEEK | MSG_TRUNC trick is not portable at
all). Although, it would be easy to set up a 64K staging buffer where
the datagram is initially copied, and then allocate a smaller buffer
and copy again)

6.       Not using PrivateKey pass phrases at all.  All private keys
must be unencrypted on the disk.
6.       Probably not fixed (Cullen)

7.       Change to DateCategory to make is more compatible with other
implementations.
7.       Not addressed (we probably should if it is easy)

13.   TLS Client Connect Inefficiency.
13.   Not addressed. Discussing on list.


Do these things need to be fixed for the 1.0 release? Can they wait
for
1.1? Opinions? (Unless these are show-stoppers, we should probably
press on)


As far as general testing goes:

make install seems to be working fine on FC4 and OS 10.4.

I have verified that "make check" works on both FC4 and OS 10.4.

tfm also works on both FC4 and OS 10.4.  (except for the
testEarlyMedia
case, but this appears to be a timing issue in the
test-case)

Scott has tested the build (at least) on Windows. I'd feel better if I
knew we were passing some test-cases.

We are passing the protos test suite on FC4 with the PEDANTIC_STACK
build flag set (ie, we fully parse every message, giving more
opportunities for the parser to explode).

It looks to me like we would pass RFC 4475 Torture-Tests (at least as
far as stack requirements go, I don't know about DUM), but until the
test-case is completely written (a LOT of work), we cannot claim to be
Torture-Tests compliant. This will have to wait.



I think we will not have any problems dropping a release candidate
today. As per Jason's suggestion, we will just be giving a tag and a
revision number for the release candidate. Unless something pops up, I
will be doing this at the end of the day (5ish Central Time).

Best regards,
Byron Campen




Attachment: smime.p7s
Description: S/MIME cryptographic signature