著者: Didier Kryn 日付: To: dng 題目: Re: [DNG] Food for thought?
Le 17/02/2024 à 18:32, Bruce Perens via Dng a écrit : > The net driver and socket interface is an artifact of coming from
> BBN's DoD project and not really being designed with the Unix paradigm.
>
> Beyond reiserfs, Hans Reiser was very big into making a database
> accessible entirely through simple files. He might have done it in
> other circumstances.
>
> But is the "Unix Way" still meaningful after 50 years? It was mostly
> built for a pipes and filters text processing paradigm that we have
> outgrown. When I think of operating system design today, the
> filesystem is an optional accessory, one completely unnecessary on
> non-server platforms, rather than the interface to all things.
>
> The file interface is also synchronous by definition. Asynchronous
> access to it doesn't look like the open/close/read/write/seek paradigm
> of Unix-style files.
>
> I think we've had enough time that we should be looking at other ways
> to do things.
I've worked with an OS where files were accessories. It's
essentially limited to industrial process control.
I consider the replacement of asynchronous signal handling and IPC
semaphores by dedicated file-like APIs is a progress. In particular, it
allows central multiplexing of events with select-like system-calls.
AFAIU BSD has implemented the same idea, even more radically than Linux
with the kqueue system-call.