fix: generate peer-id on launch

[ckerr]

Fixes #5232.

Notes: Updated torrent Peer ID generation to happen once per session, even for public torrents.

edit: the root cause of OP's bug report appears to not be a Transmission bug. The PR is still worthwhile and harmless; but following strict semver, probably should not have been merged into the bugfix line since it's not a bugfix.

[Tockman]

I must specify that my issue is about changing first, fixed part of peer_id (TR4010-XXXXXX), not the random part (TR4010-XXXXXX)!

[andreygursky]

edit: the root cause of OP's bug report appears to not be a Transmission bug.

Being curious, how this could happen. Provided Transmission sets the version in the fixed part correctly (which seems obvious) and the tracker uses exactly these 8 bytes (no other complaints), how the last could be different so often? One possible explanation: memory corruption in Transmission, which looks like has been fixed in 4.0.0.

[ckerr]

NFI. There is something that doesn't add up and I think we either have incomplete or inaccurate info from somewhere. It's not productive to speculate where that's coming from, IMO.

Just for thinking out the hypothetical -- if it were some kind of memory corruption, we would see more feedback from more users. There is something unique going on here that doesn't add up.

[ckerr]

Also if peer ID was being corrupted due to memory corruption, the tracker wouldn't be able to recognize it as Transmission

[andreygursky]

more feedback from more users

It looks like some very infrequent event (even for those "huge" 700 times it took several months), which one would only proving by running always Wireshark. Since on the other side just only one specific database design with a corresponding limitation made it visible, otherwise it went unnoticed.

the tracker wouldn't be able to recognize it as Transmission

Before writing the initial message, I've been thinking about it also. Until I realized, that the tracker got it explicitly just from "User Agent", which Transmission sets using CURLOPT_USERAGENT to "Transmission/SHORT_VERSION_STRING". This fact made me pretty sure, the tracker might indeed got garbage in the fixed part.