[Bug-tar] I: tar --listed regression

[Dmitry V. Levin]

Hi,

The listed02.at test apparently trigger regression in tar --listed.
If you run it 10000 times it may fail from few to few hundreds times
depending on how slow your host is.

I wrote a test based on listed02.at which demonstrates the problem:

i686-ext3$ time sh -c 'for i in `seq 10000`; do ../tart.sh |md5sum; done' |sort 
|uniq -c
111.02user 187.59system 4:07.39elapsed 120%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (20360409major+6334625minor)pagefaults 0swaps
     20 1318591bc0df60602e94e69623d8a98e  -
     52 6911b28b3d8d257c3a6a676e14c96085  -
      1 a639e6bf1ddaf3f2939fbaaaee3b87d4  -
   9927 fe3f12d51f1297295d935ef1f6713740  -

The problem is reproducable both with tar-1.13.25 and tar-1.15.1 with
backports from CVS, on ext3 and tmpfs filesystems, using 2.4.x and 2.6.x
linux kernels on ix86 and x86_64 arches.


-- 
ldv

tart.sh
Description: Bourne shell script

pgpeOMWOaXlhQ.pgp
Description: PGP signature