Re: [Bug-tar] [GNU tar 1.31] testsuite: 124 failed

[Chris Clayton]

Hi.

On 03/01/2019 19:50, Chris Clayton wrote:
> 
> 
> On 03/01/2019 16:33, Bruce Dubbs wrote:
>> On 01/03/2019 04:42 AM, Chris Clayton wrote:
>>> Hi,
>>>
>>> I'm building tar-1.31 on a homebrew system that is based on (Beyond)
>>> Linux From Scratch. It is very stable even though
>>> it is frequently ahead of LFS in terms of installed package
>>> versions.
>>
>> tar is not in BLFS and BLFS is not ahead of LFS.  Both are released at the 
>> same time.
> 
> Sorry, a bit imprecise with my statement there. "It" refers to my system, 
> which is sometimes ahead of LFS/BLFS.
> Additionally, I have applications installed that I don't think are in (B)LFS  
> - MATE desktop for example. It is however,
> very stable indeed.
> 
>>
>>> I am building with the latest snapshot of gcc-8.
>>>
>>> Please let me know if I there is any other information you need or
>>> can  help in any other way (e.g. test fixes).
>>>
>>> Chris Clayton
>>
>> On my reference LFS system I get:
>>
>> 208 tests were successful.
>> 21 tests were skipped.
>>
>>  15: compress: lzip                       skipped (compress-lzip.at:28)
>>  16: compress: lzop                       skipped (compress-lzop.at:28)
>>  17: compress: zstd                       skipped (compress-zstd.at:28)
>>  82: extracting symlinks to a read-only dir  skipped (extrac07.at:30)
>>  84: extracting even when . and .. are unreadable    skipped (extrac09.at:37)
>>  90: extract parent mkdir failure         skipped (extrac15.at:28)
>>  96: delay-directory-restore              skipped (extrac21.at:33)
>> 109: incremental dump when the parent directory is unreadable skipped 
>> (listed03.at:27)
>> 131: ignfail                              skipped (ignfail.at:26)
>> 150: volumes split at an extended header  skipped (multiv07.at:31)
>> 160: listing sparse files bigger than 2^33 B  skipped (sparse05.at:31)
>> 176: volume header size                   skipped (volsize.at:32)
>> 184: remove-files with compression        skipped (remfiles01.at:32)
>> 216: selinux: basic store/restore         skipped (selnx01.at:28)
>> 217: acls/selinux: special files & fifos  skipped (selacl01.at:28)
>> 224: gtarfail                             skipped (gtarfail.at:26)
>> 225: gtarfail2                            skipped (gtarfail2.at:26)
>> 226: multi-fail                           skipped (multi-fail.at:26)
>> 227: ustar-big-2g                         skipped (ustar-big-2g.at:26)
>> 228: ustar-big-8g                         skipped (ustar-big-8g.at:26)
>> 229: pax-big-10g                          skipped (pax-big-10g.at:26)
>>
>> Note that we normally build tar as root.
>>
> 
> I'll try that although my normal practice is to build everything (including 
> earlier versions of tar) as an non-root user.
> 

Building as root did not make any difference. I still get failures. The main 
thing that I've found in my investigations
over the weekend is that the results of tests 123 and 124 are not consistent 
over build/test runs.  Sometimes both fail,
sometimes one or the other fails and sometimes both succeed.

To explore further, I hacked the attached script together. Basically, it runs 
the build and tests 123 and 124 eight
times and retains testsuite.log for all runs and the directory in which each 
test's diagnostics are kept for any tests
that fail. I've run it twice and got the same inconstency. By "same", I don't 
mean that I get exactly the same instances
of successful and failed tests on each run, but I do get the same symptom of 
sometimes both fail, sometimes one or the
other fails and sometimes both succeed.

I've also tried running the tests on various kernels. I currently have 4.20.0 
and 4.19.13 installed, but I've built and
booted the latest stable kernel in the 4.14 through to 4.18 series and got the 
same variations in successes and
failures. Moreover, I've tried building tar-1.31 with gcc7, but that didn't 
help either.

As I said (or tried to say) in my original report, my system was built using 
LFS and BLFS, but now has some additional
software (e.g. MATE desktop and rpm for package build and management) and some 
packages at later versions than in the
latest release of LFS/BLFS. So, to assess whether the failures can be 
attributed to my system,I installed Linux Mint 19
onto a spare partition, installed all the available updates and ran my script 
on that. Again I get the pattern of one or
other test fails, both fail or both succeed. That, I think, suggests that my 
particular mix of software and their
versions is not the root cause of the trest failures. As further evidence of 
this, I've looked at the tar-1.31 source
rpm from the Mageai distribution and they hack tests 123 and 124 out of 
tests/testsuite.at because they fail in their
testing environment.

I'm not sure what else I can do here, other than stick with tar-1.30, but 
suggestions for additional diagnostics and/or
fixes will be very welcome. I can provide an archive of the outputs from either 
or both runs of my script, if that would
help.

>> There were no failures.  Some skips are due to missing dependencies, but 
>> that is acceptable.
>>
>>   -- Bruce Dubbs
>>      linuxfromscratch.org

Chris

rundiags
Description: Text document