Re: [Bug-datamash] datamash-1.1.0 build feedback

[Assaf Gordon]

Hello,

Thank you very much for all the testing!

On 02/10/2016 12:13 PM, Nelson H. F. Beebe wrote:
> Here are the failing output files on Solaris 10 x86-64:
>
> % cat tests/datamash-error-msgs.log
> sh: bar: not found

This is a known issue in the test file (luckily not a bug in datamash):
The shell in Solaris 5.10 treats '^' as a special character,
and the test uses 'foo^bar' as input.

I will modify the test script.

> On OpenBSD 5.0 and 5.4, I get this single failure:
>
> % cat tests/datamash-tests-2.log
> datamash-tests-2.pl: test rnd1: stdout mismatch, comparing rnd1.2 (expected) 
> and rnd1.O (actual)

This might be related to a bug in OpenBSD's ceil function,
which I reported here:
  http://marc.info/?l=openbsd-tech&m=143889843320681&w=2

Datamash should've used gnulib's replacement ( 
http://lists.gnu.org/archive/html/bug-gnulib/2015-09/msg00007.html ),
but perhaps something isn't working correctly for OpenBSD 5.4 .
It does work for newer OpenBSDs (5.7 / 5.8),
so I'll have to take a lot at installing an older OpenBSD 5.4 to debug.


> On OpenBSD 5.8, I get one failure:
> [...]
> *** 1 ****
> ! nan
> --- 1 ----
> ! -nan
>
> Note that the sign of NaN is not defined on IEEE 754 floating-point
> arithmetic; some CPUs produce negative NaNs, and others, positive
> NaNs, and I/O libraries are free to produce any sign of NaN.

I was not aware of that, thank you for the explanation.
I will change the tests accordingly.

> On NetBSD 5.0, I get one failure:
>
> % cat tests/datamash-tests.log
> datamash-tests.pl: test e17 failed: exit status mismatch:  expected 1, got 0

Will investigate.


Many thanks,
 - assaf