From: John Carter <john.carter@...>
Date: 2004-06-24T06:49:46+09:00
Subject: Re: PTY.spawn error-prone?

On Thu, 24 Jun 2004, Hal Fulton wrote:

> Ben Giddings wrote:
>> Hal Fulton wrote:
>> 
>>> I've been playing with PTY and it seems to me that this
>>> implementation is exceptionally prone to errors.

Let me elaborate a little, since I wrote this comments below....

> Allow me to quote from the rexpect docs on rubyforge:
>
>>      19   #
>>      20   # Suppose you tell it to spawn off a command that doesn't 
>> exist....
>>      21   # What do you think will happen?
>>      22   #  1) You get a sane error code?
>>      23   #  2) It raises an appropriate Exception?
>>      24   #  3) Returns quite happily and at some random time later, 
>> throws a
>>      25   #     PTY::ChildExited exception?
>>      26   #
>>      27   # Yup. Option 3) is what happens.
>>      28   # Evil isn't it?!

Firstly, most people _won't_ notice this as a problem. If you working
on an unloaded system, the PTY::ChildExited exception happens
immediately. It is when you are driving the system to the edge by
using something like rexpect to drive hundreds of processes that the
problem hits.

Looking deep into the Pty code (cvs latest) we have...

a) fork()
b) if child
c)   Do Lots of stuff
d)   exec "sh"
e)   shell does PATH lookup.
f)   PATH lookup fails
g)   shell exits "bad command"
h) ruby interpretor receives a "SIGCHLD".

Now from the moment we fork(), it is up to the OS scheduler, depending
on the current load, device(hard drive, keyboard,...) activity, phase
of moon, etc, etc how long it takes to get from a) to h).

Thus several events can get overloaded _indistinguishably_ onto that
one SIGCHLD....

    * Exec'd program couldn't be started. (Command not found,
      permission denied etc.)
    * Exec'd program failed to start. (Not a valid program)
    * Exec'd program started and died.
    * Exec'd program, started, did it's thing correctly, and exited before
      you noticed.


ie. In a sense the problem is not with Pty, it is with the notion of
invoking 'sh' to do our PATH lookups, IO redirection and command line
globbing. Then we could get much more sensible exceptions coming back
than just PTY::ChildExited

I have said it before in this forum, we need to be able to do what
"system()" and backtick and "open" and "exec" do _without_ invoking
"sh", that twenty year old downgrade to ruby.


John Carter                             Phone : (64)(3) 358 6639
Tait Electronics                        Fax   : (64)(3) 359 4632
PO Box 1645 Christchurch                Email : john.carter@tait.co.nz
New Zealand

The universe is absolutely plastered with the dashed lines exactly one
space long.