bug#14327: I am getting random hangs executing coreutils 8.21 sort.

[Chen Guo]

Hi Kevin,
On Wed, May 1, 2013 at 7:02 AM, Kevin Wills <address@hidden> wrote:
> THREAD #1
>
> (gdb) bt 20
> #0  0x0000003675c07ba5 in pthread_join () from /lib64/libpthread.so.0
> #1  0x0000000000406d92 in sortlines (lines=0x2af4898c1b50, nthreads=1, 
> total_lines=3913269, node=0x1dbf37b0, queue=0x7fffb16a4b80, tfp=0x1dbf32a0,
>     temp_output=0x1dbf323d 
> "/tmp/TEST.20100129.1367347586.5d6b2b011bb90cc1/sortrg8yHh") at 
> src/sort.c:3587
> #2  0x0000000000406d83 in sortlines (lines=0x2af48b69ccf0, nthreads=2, 
> total_lines=3913269, node=0x1dbf35b0, queue=0x7fffb16a4b80, tfp=0x1dbf32a0,
>     temp_output=0x1dbf323d 
> "/tmp/TEST.20100129.1367347586.5d6b2b011bb90cc1/sortrg8yHh") at 
> src/sort.c:3585
> #3  0x0000000000407191 in sort (files=0x6185b8, nfiles=0, output_file=0x0, 
> nthreads=4) at src/sort.c:3917
> #4  0x00000000004093a6 in main (argc=10, argv=0x7fffb16a5238) at 
> src/sort.c:4702
>

Took a quick look, noticed the above call stack is suspicious. The last call
to sortlines before the pthread_join() is invoked with nthreads = 1, while
in code sortlines would only call pthread_join() if nthreads > 1.

I confirmed this in the disassembly as well to rule out the unlikely
possibility this was the result of some compiler optimization (I used -O2).

Did you compile this yourself or was it distributed with your system?