[9fans] Pathological 9P question.

[9fans] Pathological 9P question.

[Dan Cross]

Here's a question about a pathological case.  Read(5) says that a read
in a directory must have an offset of either 0, or the offset of the
previous read plus the number of bytes returned.  Walk(5) says that a
walk to . (ie, a walk where nwnames == 0) and newfid != oldfid is
equivalent to cloning oldfid, and that both fids will point to the same
file.  What is the expected behavior, then, if a read is executed in a
directory, the fid is cloned but not clunked, and then a read is
executed on the new fid using the appropriate offset for the old fid?
Should the read succeed or fail?

	- Dan C.

Re: [9fans] Pathological 9P question.

[Charles Forsyth]

you can't walk (clone) an open fid.

Re: [9fans] Pathological 9P question.

[rob pike, esq.]

> Should the read succeed or fail?

the clone should fail. it is illegal to clone an open fid.

-rob

Re: [9fans] Pathological 9P question.

[Dan Cross]

> > Should the read succeed or fail?
>
> the clone should fail. it is illegal to clone an open fid.

Oh.  Yeah.  I, uhh...I knew that, I was just testing you.  Ahem.

Thanks to Rob and Charles for point this out.  I wonder what the
rationale for the restriction is, however; I suppose I'm thinking about
Unix dup too much when I think of cloning.  Probably an artifact of
thinking of fid's as being more like file descriptors than they really
are.  I guess the idea of duping can be adequately implemented by the
client and is unnecessary in the protocol.  Since we're minimalists,
it would therefore be illogical to include it there.

	- Dan C.

Re: [9fans] Pathological 9P question.

[rob pike, esq.]

> I wonder what the
> rationale for the restriction is

it simplifies the implementation of servers, which can make
assumptions about the uniqueness of data structures, such
as caches, associated with open files.

-rob

Re: [9fans] Pathological 9P question.

[jmk]

Version(5) is much more fun to obsess about.

Re: [9fans] Pathological 9P question.

[Dan Cross]

I've been obsessing over most of them for the last few days.  What's
*really* fun is thinking about all the concurrency issues associated
with asynchronous operations on the same files and fids and outstanding
requests....

	- Dan C.