I have some vague memory of some other piece of code that depended on that listRotateTailToHead call

I'm also worried about it, but I didn't find that code, we can merge it or wait more feedbacks?

i think we can merge it.. we both reviewed this specific concern and didn't find any problems.

i see in putClientInPendingWriteQueue, we always add the node (new client) to the head

listLinkNodeHead(server.clients_pending_write, &c->clients_pending_write_node);

and in handleClientsWithPendingWrites, we process the list from head to tail

    listRewind(server.clients_pending_write,&li);
    while((ln = listNext(&li))) {
        client *c = listNodeValue(ln);
    ...
        /* Try to write buffers to the client socket. */
        if (writeToClient(c,0) == C_ERR) continue;

        /* If after the synchronous writes above we still have data to
         * output to the client, we need to install the writable handler. */
        if (clientHasPendingReplies(c)) {
            installClientWriteHandler(c);
        }

will the order here cause problems? the new clients will be processed before old clients.
maybe it's not a problem, posted here since this seems somewhat relevant

why does it matter at which order we handle pending writes?

new clients will be processed earlier than old clients, it is not FIFO.
anyway it doesn't seem to be a problem

server.clients_pending_write is different from server.clients. When handleClientsWithPendingWrites processes server.clients_pending_write, it removes the client from the list without reinserting it, so the order is not important.