further nc_test progress

[Jianwei Li]

>When i compare it to the test.nc file created by serial netcdf, i see
>they have the same size, but different md5sums.  I don't know how big
>a deal it is, but to see the differences, run 'od' on both files and
>compare the output.

Well. I took a look at the two test.nc files. They ARE different at
a number of byteoffset with "od", but they are the same if I use "ncdump"
(an utility coming with serial netCDF) to dump their netcdf data.
I'm not quite sure what happens behind those difference, but this maybe 
caused by different underlying implentations, e.g. when extending
a record variable, they might deal with "hole"s in different ways.

>
>'nc_test -r test.nc' passes everything: as you said, there is a bit of a
>discrepancy between how we return error codes and how serial netcdf
>returns them, but its great to see "Total number of failures: 2"
>

Hmm, I got "Total number of failures: 1" :> 
i.e. "ncmpi_open of nonexistent file should have returned system error".
This would be OK.

>'nc_test test.nc' has a few problems for me:  I was getting a couple
>errors because the test did independent operations w/o putting itself
>in indep mode.  I fixed those, but there are still a few errors. Then
>it gets to "testing ncmpi_put_var_short" and just hangs.    Have you
>seen this?   I *am* building parallel-netcdf with mpich2, so maybe
>that's a factor?

Well, there are quite a lot FAILUREs in this test. 
However, I didn't get hung and get through all tests with
"Total number of failures: 914"
(Are you sure your test hangs and pops no error messages?)
I'll take a look at the test to see what happens here.

--
Jianwei