[libusb] libusb_handle_events_locked and GSource

[Richard H. <hug...@gm...>]

Hi all,

I wrote a PR for libgusb (a GObject wrapper around libusb) to
reinstate the libusb_pollfd-based integration for GSource. The link to
what I've done is here:
https://github.com/hughsie/libgusb/blob/wip/hughsie/GUsbSource/gusb/gusb-source.c
which is based on some code Hans de Goede wrote a *long* time ago.

This fixes a deadlock in fwupd when lots of threads are in flight
doing USB stuff to devices, but unexpectedly breaks fprintd. Could a
libusb_pollfd expert (if there is such a thing!) have a look at the
source file above and tell me if the locking is done correctly? Maybe
GSource doesn't quite have the semantics that libusb_lock_events is
expecting?

I've attached a gdb backtrace of fprintd if that might be helpful. I'm
not really sure that fprintf should be iterating the main context with
g_main_context_iteration() but I don't think it's the source of this
issue.

It looks like usbi_mutex_lock(ctx->open_devs_lock) seems to be
blocking -- but all the users of it look fine, and I *should* be
running from a single thread now. Ideas welcome! Thanks.

Richard.