I've looked into this problem some more and discovered that
the test actually does generate that much data due to what
looks like a bug in glibc. I've opened a bug report with
Red Hat and created an issue to track it on our end. See
https://issues.apache.org/jira/browse/STDCXX-428. I also
linked it to the infinite loop issue that Mark so speedily
opened yesterday:
https://issues.apache.org/jira/browse/STDCXX-426
Now that I know what the problem is I should be able to fix
the test so as to avoid this misbehavior.
That said, I think the problem illustrates that we should
also make exec more robust and less prone to getting into
trouble because of misbehaved tests. Maybe we should put
a cap on the amount of data tests are allowed to generate
before getting cut off? In addition, instead of reading
the test output one character at a time, exec should read
it line by line or in bigger chunks. I'll leave it to Andrew
to investigate which approach makes the most sense :)
Martin
Martin Sebor wrote:
Andrew Black wrote:
Probably wouldn't hurt to open a jira for this, as I'm currently
running a sweep of results for our other products, and this may take a
little time to look into.
Wow, looks like it's already done! (Thanks Mark!)
Interestingly,
http://people.apache.org/~sebor/stdcxx/results/redhat_as-4.0-amd64-gcc-64b-3.4.6-12d-cfg-l.gz.txt
indicates that the trunk version of the 21.cwchar test produces the
following message when run under the exec utility:
NAME STATUS WARN ASSERTS FAILED PERCNT
USER SYS REAL
21.cwchar 0 0 169 10 94% 0.000
0.000 0.020
Yeah, I've seen that too.
However, if your working copy has changed as a result of work you've
been doing, this data is likely not relevant.
That's a good point. I have reproduced the same behavior with
a fresh sync to the head of trunk.
Martin
--Andrew Black
Martin Sebor wrote:
I'm running into an (almost?) infinite loop when running some
of our tests under the exec utility on Linux (in a 12D build
with gcc 3.4.6 on Red Hat Advanced Server 4, I haven't tried
other configurations). The initial output of strace for one
of the tests, 21.cwchar, is in the attached file. The test
by itself runs fine to completion and doesn't produce any
unusual output (no NULs).
Andrew, when you have a chance, can you take a look at it?
If that's not going to be soon let me know if I should open
an issue.
Martin
