I now have completed 51 test builds of tar-1.22 on about 20 flavors of Unix in my test lab.
While about half of them have been successful, and are now installed here, there are some test failures that are repeated across platforms, making me suspect an error in the tests themselves. Here is a compact summary of the test failure reports, in ascending order of test numbers, prefixed with the frequency of occurrence in the logs: % grep -h FAIL *.log | sort | uniq -c 1 2: decompressing from stdin FAILED (pipe.at:33) 6 22: extracting selected members from pax FAILED (extrac05.at:38) 1 23: mode of extracted directories FAILED (extrac06.at:36) 1 26: incremental FAILED (incremental.at:26) 1 27: restore broken symlinks from incremental FAILED (incr01.at:30) 4 30: working --listed FAILED (listed02.at:31) 1 31: renamed files in incrementals FAILED (incr03.at:31) 1 32: proper icontents initialization FAILED (incr04.at:34) 3 33: renamed dirs in incrementals FAILED (rename01.at:27) 3 34: move between hierarchies FAILED (rename02.at:27) 3 35: cyclic renames FAILED (rename03.at:26) 1 36: renamed directory containing subdirectories FAILED (rename04.at:30) 2 37: renamed subdirectories FAILED (rename05.at:27) 1 40: link count gt 2 FAILED (link01.at:36) 1 46: multivolume dumps from pipes FAILED (multiv01.at:30) 1 49: split directory members in a MV archive FAILED (multiv04.at:39) 10 67: compressed format recognition FAILED (comprec.at:25) 1 70: truncate FAILED (truncate.at:32) In each case, I have refrained from installing the newly-built tar. Notice test #67: I see that failure on GNU/Linux (Alpha and IA-32), FreeBSD 6 (IA-32), Mac OS X (PowerPC and IA-32/EM64T/AMD64), NetBSD (IA-32), and Solaris x86. Test #22 fails only on Mac OS X (PowerPC and IA-32/EM64T/AMD64). Test #30 fails on Mac OS X (PowerPC only), OSF/1 Alpha, Solaris (only IA-32, not AMD64). Tests #33--37 fail on FreeBSD 6, Mac OS X (PowerPC only), and OSF/1 Alpha. I can provide build logs on request, but perhaps reports from others will show whether my test failures are common and widespread. On some platforms, I can do additional builds with several other compilers (nvcc, opencc, pathcc, pgcc, suncc, ...), to weed out possible compiler code-generation errors that might be responsible for test failures. ------------------------------------------------------------------------------- - Nelson H. F. Beebe Tel: +1 801 581 5254 - - University of Utah FAX: +1 801 581 4148 - - Department of Mathematics, 110 LCB Internet e-mail: [email protected] - - 155 S 1400 E RM 233 [email protected] [email protected] - - Salt Lake City, UT 84112-0090, USA URL: http://www.math.utah.edu/~beebe/ - -------------------------------------------------------------------------------
