On Feb 21, 2017 11:55 AM, "Joerg Schilling" <schily@schily.net> wrote:
Random832 <random832@fastmail.com> wrote:

> Why does that make them more useful for end users than device filenames,
> especially for non-SCSI devices? USB or ATA bus numbers - or USB device
> identifiers - might be more useful than SCSI bus numbers, for end users
> who have USB or ATA devices. ATA bus numbers had a deterministic mapping
> to /dev/hd* device filenames, once upon a time.

You know that Linux implements several different competing methods to access
these drive types?

You know that some of them do not (or not correctly) support DMA?

You know that DVD drives will not work for writing if you do not have DMA?

All of which *could* be supported in a kernel driver.

> Why does that mean the correct solution is "require the user to type in
> the bus number on a program's command line" rather than "configure a
> particular bus number to have a particular filename"?

Why do people always repeat the same questions that have been answered many
times in the past already?

I don't think it was a question. It was a philosophical statement.

1)      CAM is a standard, why not use it? full stop

2)      **Most** Operating systems do not support /dev/* based access to SCSI.
        This includes a POSIX certified system like Mac OS X.

3)      **Most** Operating systems do not even support a file descriptor based
        interface to SCSI commands.
        This includes a POSIX certified system like Mac OS X.

4)      CAM is the only interface method that can be mapped to all
        existing operating system specific implementations. 

In other words, you worked around perceived problems in what most on this list would consider the traditional file based UNIX method by implementing in terms of a different standard. That's fine, of course, but is independent of the philosophical point that it is not in keeping with the traditional method.

It's perhaps noteworthy that writing to a CD or DVD on Plan 9 is done with cat (and networking is implemented in terms of opening, reading and writing named files). So it's certainly *possible* to use a filesystem method to write physical media, even if impractical on the array of real-world systems you want to support.

        - Dan C.