Re: [Bug-tar] Problem with fstatat on AIX 7.1

[Paul Eggert]

On 09/03/11 08:31, Kevin Brott wrote:
> So this bundle seems to have the fstatat issue resolved.

OK, well, we've made progress but we may not be out of the
fstatat woods let.  Let's turn to the next test failure.
Your earlier message gave these symptoms:

--- /dev/null   2011-09-01 14:53:45 -0700
+++ /lppdir/build/phs-gtar/phs-gtar/tests/testsuite.dir/at-groups/38/stderr     
2011-09-01 14:53:46 -0700
@@ -0,0 +1,2 @@
+tar: .: Cannot stat: Invalid argument
+tar: Exiting with failure status due to previous errors
./extrac12.at:26: exit code was 2, expected 0
38. extrac12.at:23: 38. extract dot permissions (extrac12.at:23): FAILED 
(extrac12.at:26)

Is this still what testsuite.log is reporting?  If so, here's
the failing test:

  TEST_TAR_FORMAT=gnu
  export TEST_TAR_FORMAT
  TAR_OPTIONS="-H gnu"
  export TAR_OPTIONS
  rm -rf *

  mkdir src dst
  echo file1 >src/file1
  echo file2 >src/file2
  chmod a-w src

  tar --no-recursion -cf archive.tar -C src . ./file1 file2 &&
  tar -xf archive.tar -C dst ...

Can you please try this test by hand, with your tar, but use
"truss" to save the output of the two tar invocations?  Something
like this:

  truss -o tr1 tar --no-recursion -cf archive.tar -C src . ./file1 file2 &&
  truss -o tr2 tar -xf archive.tar -C dst ...

This should isolate which "tar" is failing, and give us a clue as to why.

PS.  Have you reported the bug with fstatat (AT_FDCWD, "conftest.file", &a, 0)
to the AIX folks?  This time it's clearly a bug; the earlier problem is arguably
a problem with tar rather than with AIX.