Re: info/t/dir-dangling-entry.sh hang on netbsd

[Gavin Smith]

On 1 June 2015 at 01:15, Gavin Smith <address@hidden> wrote:
> On 1 June 2015 at 00:52, Karl Berry <address@hidden> wrote:
>> Hi Gavin - on a netbsd 5.1 system I happen to have access to, the
>> dir-dangling-entry test hangs (in uninterruptible disk wait; parent
>> process had to be killed to recover).  Running with sh -vx shows it
>> hanging in Init-inter.inc:
>> ..
>> # Wedge open pipe
>> sleep 10 3>$PTY_TYPE &
>> + sleep 10
>> [no more output]
>
> It's probably freezing on trying to open the FIFO. I don't have
> anything to suggest at the moment.

I haven't tried it on netbsd yet, but I suspect there is a difference
in opening a FIFO with GNU/Linux, either with Linux, or with how the
shell does it (e.g. what flags it uses for the system calls). I have
not been able to get an uninterruptable process status (that's "D"
with ps) by doing "mkfifo tmp.fifo" followed by "sleep 1 >tmp.fifo &"
or similar: it's always been "S", interruptable sleep.

Anyway, I've attached a patch for a different approach. It works on my
system; hopefully it will work on other systems too. Can anyone try
it? The approach is to pass the name of the FIFO file as a
command-line argument to pseudotty, which means the FIFO can stay
closed for longer and we shouldn't have to mess around with "wedges"
to keep the FIFO open for reading/writing.

As an aside, this stuff is a lot harder than I thought it would be
when I started. I thought it could be a cool afternoon hack! Everyone
says that Un*x is great because you can pipe programs together, but
this seems only to be easy when they are in a straight line and in a
single direction. When you have multiple pipes going in both
directions, or make your pipes loop-the-loop, it seems a lot harder to
get right.

pass-fifo-to-pseudotty.diff
Description: Text document