I'll also make a test to move completed handles over to a new list so that they can be skipped in the main loop as well.

The pytest failures are very curious. I need to dig into those...

@icing I could use a little hand-holding to learn the best way to debug these pytest failures.

@icing I could use a little hand-holding to learn the best way to debug these pytest failures.

I'll checkout this branch and build locally in a few minutes.

I see problems locally, from runs stalling for a second to busy loops. The CI failures look like busy loops where the started curl simply does not return.

I have not pinned down the cause. But the order in which transfers are started is no longer the same as before. This stems from the fact that PENDING transfer are no longer processed at all. Simple example with 3 transfers:

start: 1-CONNECT, 2-CONNECT, 3-CONNECT
process 1: 1-CONNECTING, 2-CONNECT, 3-CONNECT
process 2: 1-CONNECTING, 2-PENDING, 3-CONNECT
2 is moved to pending and removed from easy list
           1-CONNECTING, 3-CONNECT
process 1: 1-PROTOCONNECT, 3-CONNECT
multi changed and pending is inspected
           1-PROTOCONNECT, 3-CONNECT, 2-CONNECT
process 1: 1-DO, 3-CONNECT, 2-CONNECT
process 1: 1-DID, 3-CONNECT, 2-CONNECT
process 1: 1-PERFORMING, 3-CONNECT, 2-CONNECT
process 1: 1-DONE, 3-CONNECT, 2-CONNECT
process 3: goes through all states
*** stalled for 1 second **
process 2: goes through all states

I think the problem is that you modify mult->easyp while iterating over it.

if(data) {
  do {
    run_single(data);   /* this may modify the list */
    data = data->next;
  } while(data);
}   

I think the problem is that you modify mult->easyp while iterating over it.

Ah yes indeed.

@icing thanks for that hint, it seems that was the bug that held this back.

@icing thanks for that hint, it seems that was the bug that held this back.

Yep, works now locally here as well.