There was some bug which could be worked around by double emitting an event.
However that proofed to be brittle:

We send the first message here from the worker goroutine:
https://git.sr.ht/~sircmpwn/aerc/tree/a5553438/item/worker/maildir/worker.g=
o#L306

Then Tick() is waked in the main goroutine and calls ProcessMessage:
https://git.sr.ht/~sircmpwn/aerc/tree/a5553438/item/widgets/account.go#L100

ProcessMessage in the main goroutine reads types.Message state with
msg.getId() and msg.InResponseTo():
https://git.sr.ht/~sircmpwn/aerc/tree/a5553438/item/worker/types/worker.go#=
L74-76

Meanwhile in the worker goroutine we call PostMessage for a second
time with a pointer that points to the *same* previous message that
ProcessMessage is reading:
https://git.sr.ht/~sircmpwn/aerc/tree/a5553438/item/worker/maildir/worker.g=
o#L306

The second PostMessage call makes writes to message while
ProcessMessage in the main goroutine is possibly reading:
https://git.sr.ht/~sircmpwn/aerc/tree/a5553438/item/worker/types/worker.go#=
L59

This led to a data race in the event loop

Reported-By: Wagner Riffel <w@104d.net>

Trims whitespace in list of excluded notmuch tags. This allows a comma
separated list with spaces to be correctly processed.

We frequently had issues with notmuch segfaulting and my guess is that this
was due to the garbage collection magic used in the module.

This changes to a fork that ripped the functionality out.

We made a new type out of go-message/mail.Address without any real reason.
This suddenly made it necessary to convert from one to the other without actually
having any benefit whatsoever.
This commit gets rid of the additional type