dev_t / udev_t confusion ?

[Poul-Henning Kamp]

I have an item on my TODO list which says "fix dev_t / udev_t kernel
confusion before 5-STABLE ?".

The confusions is that in userland dev_t is an integer type which
encodes the major+minor number of a device, in the kernel it it a
pointer to "struct cdev" which represents the device to the kernel.

Back when dev_t became a struct pointer, I counted the number of
kernel source files which used the kernel dev_t vs the ones that used
the useland dev_t and found something like 325:25.

Since we share a number of device drivers with other OSs, and
following the simple plurality, I called the userland dev_t "udev_t"
in the kernel.  And therefore, "dev_t" in userland and "dev_t" in
the kernel are entirely different.

We had a discussion about fixing this some point back, and I am not
sure if we really reached closure on it.

The change proposed is more or less to do:
	s/dev_t/struct cdev */
	s/udev_t/dev_t/
over all the kernel sources (366 files or so).

The benefit is that we get the dev_t/udev_t confusion solved, the
disadvantage (apart from the churn) is that we reduce the already
limited direct source compatibility with other BSDs a bit further.

Personally I'm pretty 50/50 on this issue, but if we want to do it,
we want to do it before 5-STABLE, not after (to avoid the FreeBSD3
syndrome).

Personally I don't think there is much need for a long discussion
and I would prefer to see simply a show of hands for yes and no,
and any hear any really heavy duty arguments pro et contra. 


-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.