[core] make sure TTL will not drop packets over last block

[gou4shi1]

The msg seq of m_pLastBlock is undefined, right?

[gou4shi1]

any comment?

[maxsharabayko]

Your improvement looks good.
However, seems like another loop at the beginning of the function is also not very bullet-proof: no checks for p == NULL and for m_pLastBlock.

    for (int i = 0; i < offset; ++i)
        p = p->m_pNext;

[gou4shi1]

I have check for this loop, although return -2 is ugly :lol:

[ethouris]

As long as you add the explanation of the return values in the function's documentation, there's no problem with it.

[gou4shi1]

I have just checked the code in buffer.cpp/h, p should never be NULL, it's a ring buffer, right?

[maxsharabayko]

I have just checked the code in buffer.cpp/h, p should never be NULL, it's a ring buffer, right?

Yes. Packets are added sequentially to the buffer, and there should be no gaps.
See CUDT::sendmsg2.

[ethouris]

Originally they didn't even have sequence numbers until they were sent out for the first time. This has changed only with the group code, and it's only for synchronization. There shouldn't ever happen a case of existing a gap in the sender buffer, even in the group code, in which case there may happen an "initial shift" due to link activation that causes a difference between ISN and the first packet sequence, but this is resolved by wiping out this initial part of the buffer and causing an artificial loss that is ignored at the receiver side.

[ethouris]

Something's wrong. Can you please merge against the latest master ?

[gou4shi1]

I don't known why add a log will make the unit test failed... it success in my local

[gou4shi1]

thx for the merge :)

[maxsharabayko]

@gou4shi1 Thank you for the contribution! 😄