[9fans] rio snarf

[9fans] rio snarf

[Fco. J. Ballesteros]

Any reason why it updates on close rather than truncating on open(OTRUNC)?

Re: [9fans] rio snarf

[Russ Cox]

> Any reason why it updates on close rather than truncating on open(OTRUNC)?

Surely you have a different question.

Russ

Re: [9fans] rio snarf

[Fco. J. Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 580 bytes --]

Well. We just found that removing snarf from rio fsys makes things
work great wrt sharing the snarf buffer across machines.

At first, we thought that if was enough to call open(OTRUNC) while
copying to the snarf buffer, then we noticed the deadlock with the
open("/dev/snarf") in rio, then we tried removing snarf from rio, and,
voila.

Either I'm crazy because of the lots of coffee I have been drinking, or
there is something I'm missing regarding why rio has snarf builtin instead
of using a regular file for that.

Hope my question was more clear now.

cheers

[-- Attachment #2: Type: message/rfc822, Size: 2984 bytes --]

From: Russ Cox <russcox@gmail.com>
To: Fans of the OS Plan 9 from Bell Labs <9fans@cse.psu.edu>
Subject: Re: [9fans] rio snarf
Date: Wed, 27 Oct 2004 13:10:45 -0400
Message-ID: <ee9e417a0410271010440c7d9b@mail.gmail.com>

> Any reason why it updates on close rather than truncating on open(OTRUNC)?

Surely you have a different question.

Russ

Re: [9fans] rio snarf

[Russ Cox]

> Well. We just found that removing snarf from rio fsys makes things
> work great wrt sharing the snarf buffer across machines.

If that's all you want to do, just create a /dev/snarf 
before rio runs -- if it sees one already there, it won't
provide one.

> At first, we thought that if was enough to call open(OTRUNC) while
> copying to the snarf buffer, then we noticed the deadlock with the
> open("/dev/snarf") in rio, then we tried removing snarf from rio, and,
> voila.

This I still don't fully understand.  The reason the snarf buffer doesn't
change until close is probably lost to history, but one nice effect
is that updating the snarf buffer is atomic -- you'll never see a 
partially written one.

> Either I'm crazy because of the lots of coffee I have been drinking, or
> there is something I'm missing regarding why rio has snarf builtin instead
> of using a regular file for that.

Rio used to provide a separate snarf buffer for each instance of rio.
It's only recently that there has been this push toward having one
snarf buffer for the whole system.

Russ

Re: [9fans] rio snarf

[Fco. J. Ballesteros]

[-- Attachment #1: Type: text/plain, Size: 451 bytes --]

Well, having a /dev/snarf created by, say,
binding /tmp/blah to /dev/snarf
does not seem to work, because rio ends up
leaving the biggest thing that was in there
(because it does not truncate the file).

I think it assumes that the behaviour of /dev/snarf is that
of rio's, and not that of a regular file.

Regarding the question, I think you answered it. I understand it
provides a snarf if you have different per-rio snarf files.

thanks

[-- Attachment #2: Type: message/rfc822, Size: 4005 bytes --]

From: Russ Cox <russcox@gmail.com>
To: Fans of the OS Plan 9 from Bell Labs <9fans@cse.psu.edu>
Subject: Re: [9fans] rio snarf
Date: Wed, 27 Oct 2004 13:51:10 -0400
Message-ID: <ee9e417a04102710516945fe9e@mail.gmail.com>

> Well. We just found that removing snarf from rio fsys makes things
> work great wrt sharing the snarf buffer across machines.

If that's all you want to do, just create a /dev/snarf 
before rio runs -- if it sees one already there, it won't
provide one.

> At first, we thought that if was enough to call open(OTRUNC) while
> copying to the snarf buffer, then we noticed the deadlock with the
> open("/dev/snarf") in rio, then we tried removing snarf from rio, and,
> voila.

This I still don't fully understand.  The reason the snarf buffer doesn't
change until close is probably lost to history, but one nice effect
is that updating the snarf buffer is atomic -- you'll never see a 
partially written one.

> Either I'm crazy because of the lots of coffee I have been drinking, or
> there is something I'm missing regarding why rio has snarf builtin instead
> of using a regular file for that.

Rio used to provide a separate snarf buffer for each instance of rio.
It's only recently that there has been this push toward having one
snarf buffer for the whole system.

Russ