libusual interface [was Re: libusual and ub]

[Pete Zaitcev]

On Mon, 19 Dec 2005 11:40:58 -0500, Bill Nottingham <notting at redhat.com> wrote:

> > > I mean separate the aliases *into the base drivers themselves* - not
> > > using libusual at all. i.e., a given usb id only matches one of
> > > usb-storage or ub.
> > 
> > Nothing new in this approach either. This is how ub worked way before
> > libusual, and before Adrian Bunk broke it. The problem is, users require
> > a capability to switch between drivers. And we cannot predict which one
> > they want to use for any given device.
> 
> *Why* do they need the capability? Users *should not care* which
> driver they use - if they do, one of them needs fixed.

Well, such is the theory, but in practice making an alternative driver
to handle all cases creates an impossibly high barrier to entry.
The common solution is to pass known-good devices from old driver to
new driver (e.g. 8139too, e100). This is, unfortunately, impossible
in case of USB storage.

To answer your specific question, when Bunk was making trouble, ub did
not work as well as it does now, so there was an implicit assumption
that usb-storage would be superior under some, hitherto unknown,
circumstances. Some people were assumed to desire our SCSI stack and
old device names. Also, without libusual (not existing back then),
there was no exception for USB connected ATAPI tapes. So we did not
discuss specific instances of breakage much, and I did not challenge
him to produce failure scenarios.

-- Pete