Linux fixes for dispatch-806 merge

[dgrove-oss]

Collection of small fixes to dispatch-806 merge to
allow code to compile/link/run on Linux.

[dgrove-oss]

Not intending this as the final fix for this code; just an expedient hack to allow the merge to proceed.

[MadCoder]

this diff is wrong, it should always use DISPATCH_WLH_GLOBAL

[dgrove-oss]

ok; I think I should then add an assert that the value of the wlh parameter is DISPATCH_WLH_GLOBAL. I don't like ignoring the value the caller passed in without asserting it has a known value.

[MadCoder]

no it needs to be GLOBAL here, event_kevent.c does the same. it is meant to be that.

[dgrove-oss]

Maybe we're looking at different pieces of code. In event_kevent.c, I think the matching piece of code is:

bool
_dispatch_unote_register(dispatch_unote_t du, dispatch_wlh_t wlh,
		dispatch_priority_t pri)
{
	dispatch_assert(!_dispatch_unote_registered(du));
	du._du->du_priority = pri;
	switch (du._du->du_filter) {
	case DISPATCH_EVFILT_CUSTOM_ADD:
	case DISPATCH_EVFILT_CUSTOM_OR:
	case DISPATCH_EVFILT_CUSTOM_REPLACE:
		du._du->du_wlh = wlh;
		return true;
	}
	if (!du._du->du_is_direct) {
		return _dispatch_unote_register_muxed(du, DISPATCH_WLH_GLOBAL);
	}
	return _dispatch_kq_unote_update(wlh, du, EV_ADD | EV_ENABLE);
}
```
``

[das]

no, event_kevent.c also sets it to 'wlh'. I think these are arguably both wrong and should both be the GLOBAL constant (but it also doesn't really matter for this 'synthetic event' case)

[MadCoder]

everything makes perfect sense to me but for the commented hunk.

[das]

should this really be defined at all ? I would have thought the #ifdef EV_UDATA_SPECIFIC below would have done the right thing if it is not

[dgrove-oss]

My recollection is that @MadCoder preferred us to define it as 0x0000 instead of leaving it undefined. I believe it is a matter of style; I think either leaving it undefined or defining it to be 0x0000 accomplish the same thing.

[das]

the conditional below defines it to 0x0000 already, that is what I am talking about

[das]

i.e.

#ifdef EV_UDATA_SPECIFIC
#	define DISPATCH_EV_DIRECT		(EV_UDATA_SPECIFIC|EV_DISPATCH)
#else
#	define DISPATCH_EV_DIRECT		0x0000
#	define EV_UDATA_SPECIFIC		0x0000
#	undef  EV_VANISHED
#	define EV_VANISHED				0x0000
#endif

that is there exactly for the cases where EV_UDATA_SPECIFIC isn't present (in both cases of backend, we also support the absence of this flag with kevent)

[das]

(and in particular doing it that way will also cause EV_VANISHED to be 0x0000 on Linux. currently it is 0x0200 via the #define at the top of the file)

[dgrove-oss]

ok. makes sense.

I changed to not define EV_UDATA_SPECIFIC on Linux and squashed/rebased.

[das]

cool, lgtm, thanks!

let's not worry about this inconsistency for this review, matching event_kevent.c seems fine

[MadCoder]

Indeed, seems I misremembered, this is fine :)