[9fans] i say, this is interesting

[9fans] i say, this is interesting

[Sam]

I use rob's ``plumb to web'' rule posted
a while back.  Today I plumbed and got
a really neat result: all of my terminals
started ignoring every other character
input.

I would have found this to be a great
nuisance if it hadn't eventually gone away.
I played with it a bit, and I think I
know the causes for both the start and
stop of the feature.

A while back I upgraded from Opera 6.12
to 7.11.  I guess it's better, though I
don't really perceive any additional
functionality and it's doggedly slower.
In any event, apparently -newpage no
longer works properly because instead
of opening a new tab, it kicks off a new
opera.

The aforementioned terminal feature
persists until a second or so after
I close the newly opened opera window,
indicating the problem might be due
to the blocked ssh connection.  I do
believe that under 6.12 ``opera -newpage''
returned immediately, having delivered
the open page command to the running
instance.

Now, is this funny a bug in source or
could it be in my setup?  I have the most
amazing ability to step on things
inadvertantly causing updates to not
get applied, so the latter wouldn't
suprise me.

Cheers,

Sam

Re: [9fans] i say, this is interesting

[Sam]

<sigh>

I am a boob.  -newpage works fine, Opera just
isn't smart enough to figure out that when
I log in as sah I really want him to send
the open to the opera running as root.  I can't
blame him.

Doubly boobalicious, I should have seen that
redirecting the stdin for plumb start's ssh would
have suppressed this.

Thanks,

Sam

Re: [9fans] i say, this is interesting

[Aki M Nyrhinen]

On Fri, Jun 27, 2003 at 09:21:14AM -0400, Sam wrote:
> I use rob's ``plumb to web'' rule posted
> a while back.  Today I plumbed and got
> a really neat result: all of my terminals
> started ignoring every other character
> input.

You can get rid of this nice feature by giving
/dev/null as standard input for ssh (ie. ssh foo 'whatever'
</dev/null). I don't want to make guesses why exactly every
second character is lost (other's know, for sure), but it
seems this isn't neccessarily a bug at all.

	-Aki

Re: [9fans] i say, this is interesting

[rob pike, esq.]

i've seen that too, and i think you're right that it's a blocked ssh.
the obvious fix would be to redirect the stdin of the ssh, but i haven't
tried it.  maybe i'll do that now.  just put
	</dev/null
on the end of the 'start' line.

in the meantime, thanks for the heads up about opera -newpage.
it would be a real shame if there isn't a fix. perhaps -newtab?
i learned about newpage by reading the usage message, i think.

Re: [9fans] i say, this is interesting

[rob pike, esq.]

> I don't want to make guesses why exactly every
> second character is lost (other's know, for sure), but it
> seems this isn't neccessarily a bug at all.

oh, that's easy.  two processes reading the same file,
each one byte at a time.  what would be the result of
your kernel implementation under those circumstances?
ours just gates each process through the queue, and
the result is alternate characters going to alternate
processes.

Re: [9fans] i say, this is interesting

[Aki M Nyrhinen]

On Fri, Jun 27, 2003 at 08:24:30AM -0700, rob pike, esq. wrote:
> > I don't want to make guesses why exactly every
> > second character is lost (other's know, for sure), but it
> > seems this isn't neccessarily a bug at all.
> oh, that's easy.  two processes reading the same file,
> each one byte at a time.  what would be the result of
> your kernel implementation under those circumstances?

An impossible to find non-deterministic malfunction in the future,
of course.

> ours just gates each process through the queue, and
> the result is alternate characters going to alternate
> processes.

Now that you said it, it's very obvious because the keyboard
gives characters one at a time.

Is there a reason why the plumber does not close(0)?

	-Aki

Re: [9fans] i say, this is interesting

[rob pike, esq.]

> Is there a reason why the plumber does not close(0)?

not that i can think of. it probably should, although i don't
remember how it builds the subprocess.

-rob