Yes, I had in mind the tcp_rel_seq_table hash. I don't think you want to add one for each dissector dealing with their own conversations... If the file is read sequentially at least one time, a bsearch() or similar is usable, you insert at the end and don't have to sort .Yes, but you also have to be able to make sure you get the right answer *after* the initial sequential pass through the capture - i.e., if you click on some random frame, and the dissector for some protocol in that frame looks for the conversation to which the frame belongs, it has to find the *correct* conversation, regardless of which frames the user clicked on before clicking on that frame.
Didier
