Re: [PATCH] RT: epca_lock to DEFINE_SPINLOCK

[Nikita Danilov <nikita@clusterfs.com>]

Roland Dreier writes:
 >     Arjan> this is really ugly though; at minimum a DEFINE_STATIC_SPINLOCK()
 >     Arjan>  would be needed to make this less ugly.
 > 
 > huh?  This is a totally standard kernel idiom -- just do
 > 
 >     grep -Er 'static (DECLARE|DEFINE)' .
 > 
 > in a kernel tree to see how prevalent it is.

It may be widely used and still ugly. The general problem with
DEFINE_FOO() macros is that they obfuscate things: they do not _look_
like C variable declarations, and, in particular, type of variable is
not immediately obvious.

The only reasonable case where DEFINE_FOO(x) is really necessary is when
initializer uses address of x, but even in that case something like

        spinlock_t guard = SPINLOCK_UNLOCKED(guard);

is much more readable than

        DEFINE_SPIN_LOCK(guard);

The question is: does RT really have to force DEFINE_* as the only way
to define things?

 > 
 >  - R.

Nikita.