Philippe Gerum wrote:
Jan Kiszka wrote:
Paolo Mantegazza wrote:
...
So I think you are right in saying that one should distinguish
between cli/sti used either for accessing the hardware or for fast
atomic stuff. In the first case there _must_ be rtai_hw_something or,
since adeos is there anyhow, adeso_hw_something directly.
In the fast atomic case the fix of Jan is OK. So it is up to him to
know what he's using the oldnames for.
RTnet just needs protection against IRQs for the RTAI domain, if hard
or "adeos-optimistic" doesn't matter.
In any case the patch is due as the name rtai_local_something are no
more in 3.2, so their use will crash the simple making. In fact in
3.2 I reversed to the naming I liked since DOS times, as
local_irq_save means just saving the irq state flag not clearing it
too. So I prefer to let it for Linux magicians and use a naming that
says it long and clear and thus not hide what's behind to poor people
like me.
Actually, I considered dropping oldnames and turning to the official
API. But as 3.1 and 3.2 differs in this regard and detecting 3.2 is
not yet reliable, I decided to fix the old API first.
Actually, oldnames was there to provide temporary backward compat before
we eventually get rid of hard_xxx() stuff in the codebase in 3.x. I've
initiated this change for naming consistency purpose, and to avoid
further namespace conflicts with other non-RTAI kernel components, by
using rtai_*prefix everywhere. Since Paolo reverts to the DOS naming way
and is bringing back hard_xxx again in the 3.2 codebase, rtai_oldnames.h
is becoming useless.
A bit of confusion here. I'm reverting to the names I like but is not
the hard_xxx stuff anyhow. In fact there should be no hard_xxx stuff in
RTAI 3.2. The hard_xx need a redifining macros in oldnames.h, as in 3.1,
and rtai_oldnames must be maintained consistent for all users that keep
that do not want to change their legacy code.
Jan's request and fix proves the point.
Paolo.
