NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/comp/unix/sysv386/13816

Text File | 1992-08-30 | 1.8 KB | 39 lines

Newsgroups: comp.unix.sysv386
Path: sparky!uunet!unislc!dold
From: [e-mail] (Clarence Dold)
Subject: Re: cpio one file at a time?
References: <[e-mail]>
Message-ID: <[e-mail]>
Organization: Unisys Corporation SLC
Date: Mon, 31 Aug 1992 01:19:29 GMT
Lines: 28

From article <[e-mail]>, by [e-mail] (Chris Sells):

> The 'mv' was only an example. I'll be performing multiple operations on
> each file and I need to be able to do it before the next file is loaded.
> BTW, what's pax?

Methinks that a named pipe fits into this scheme somewhere.
I might be inclined to read directly from the tape to a file, perhaps via my
code, or dd, where the file on disk is a named pipe that would be a raw
cpio dataset.  Then the "special" process would read from the raw data, and
pipe to cpio.  While monitoring the files that cpio is creating and filling
(by some undescribed method), the "special" process would stop feeding data
down the pipe to cpio until the manipulation was done.  Then the data feed
would resume into cpio.
The named pipe would be some system determined maximum size on disk, I think
5K.  The tape coming in would stop while the pipe was at its maximum size,
and feed when the named pipe is drained.  The "special" process would act as
a throttle, controlling the data rate into cpio.  Without figuring out how
to monitor the internals of a cpio archive, I might do something as crude as
a cpio -it to find the file names on disk, storing the names, then start in
earnest, watching for each to be created (or modified), as an indication that
the previous one was fully restored.
Named pipes are created with "mknod foobar p"

--
---
Clarence A Dold - [e-mail]
               ...pyramid!ctnews!tsmiti!dold