[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Fix for back-shell child process deadlocks (ITS#2262)

At 02:20 PM 1/9/2003, simon.brady@otago.ac.nz wrote:
>On Thu, 9 Jan 2003, Kurt D. Zeilenga wrote:
>> There is one major caveat... such "fixes" are inherently non-portable.
>Could you please expand on this?

Basically, as POSIX did not define the semantics of pthreads
and fork(), the behavior of fork() in pthread'ed applications
is operating system/library system dependent.   Not only
will your fix not solve the problem on all systems, it likely
will cause some "working" systems to fail.

>I was about to respond that back-shell has filled a very important niche
>for us in production services, but I guess you're saying we should be
>using a new, smarter backend like back-perl or back-meta?

I have not said, nor do I intend to say, what software you
should be using and how.  That's your choice.

I am only attempting to communicate that, in my opinion, it is
unlikely that your patch will not be incorporated and why.

>That makes
>sense, but I suspect I'm far from the only one using back-shell beyond
>what it was intended for - I can understand how you'd rather be putting
>your resources into the new backends, but I still feel this fix is good 
>for people not ready to transition to 2.1.

I think the issue here is not really a matter of where I, you,
or others might put their resources, because if it a portable
generally-useful fix was developed it likely would be committed.
We, in fact, committed one potential fix in an earlier version
of 2.1.  It was found to be problematic and was backed out.