make check fails with error message "wrong values"

[Jens Doleschal]

Hi,

during the build of abinit I run make check and for etsf i/o it fails at some points:

== | check 'atom_species_names' values Failed ==

     ***
     *** ETSF I/O ERROR
     ***
     *** Backtrace : check_mem_str_1D()
     *** Action performed : specifications variable
     *** Target (name) : atom_species_names
     *** Target (id) : -1
     *** Error message : wrong values
     ***

== | read 'chemical_symbols' values OK ==
== | check 'chemical_symbols' values Failed ==

     ***
     *** ETSF I/O ERROR
     ***
     *** Backtrace : check_mem_str_1D()
     *** Action performed : specifications variable
     *** Target (name) : chemical_symbols
     *** Target (id) : -1
     *** Error message : wrong values
     ***

== | read 'pseudopotential_types' values OK ==
== | check 'pseudopotential_types' values Failed ==

     ***
     *** ETSF I/O ERROR
     ***
     *** Backtrace : check_mem_str_1D()
     *** Action performed : specifications variable
     *** Target (name) : pseudopotential_types
     *** Target (id) : -1
     *** Error message : wrong values

Any ideas?

Best regards, Jens

[Damien (damien-caliste)]

Hello,

Le 03/05/2012, Jens Doleschal <email address hidden> a écrit :
> Hi,
>
> during the build of abinit I run make check and for etsf i/o it fails
> at some points:
Ok, that's interesting and need correction. May I ask which compiler
are you using (I guess which version of GFortran is more accurate) ?
I'll try to reproduce and provide a correction.

Damien.

[Jens Doleschal (jens-doleschal)]

Hi Damien,

I used the open64 compiler in version 4.5.1 with the following flags:
-O3 -mtune=bdver1 -mso

Best regards, Jens

On May 16, 2012, at 10:43 AM, Damien Caliste wrote:

> Hello,
>
> Le 03/05/2012, Jens Doleschal <email address hidden> a écrit :
>> Hi,
>>
>> during the build of abinit I run make check and for etsf i/o it fails
>> at some points:
> Ok, that's interesting and need correction. May I ask which compiler
> are you using (I guess which version of GFortran is more accurate) ?
> I'll try to reproduce and provide a correction.
>
> Damien.

--
Dipl.-Math. Jens Doleschal
Wissenschaftlicher Mitarbeiter

Technische Universität Dresden
Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH)
(Center for Information Services and High Performance Computing (ZIH))
Interdisziplinäre Anwenderunterstützung und Koordination (IAK)
(Interdisciplinary Application Development and Coordination (IAK))

Willersbau A109
Zellescher Weg 12
01062 Dresden

phone [phone]
fax [phone]
email <email address hidden>

[David Strubbe (dstrubbe)]

I would like to report that I found the same failures for 1.0.3 and 1.0.4 for the "Create file test_write_geometry.nc" and "Create file test_copy_geometry.nc" sections of the test, using g95 0.93 on MacOS with "-O3", and also with "-g -O0".

[Nick Papior (nickpapior)]

Just as David Strubbe, I get the exact same errors on my setup.
It seems like it is due to a too big string array in the tests_module. The passed variable is in consecutive memory, which doesn't follow the readed values.

If I correct the routine interface:
subroutine check_read_nD(ncid, varname, type, dims, lstat, error_data)
    ...
    character(len = 256), allocatable :: strings(:)

to:

subroutine check_read_nD(ncid, varname, type, dims, lstat, error_data)
    ...
    character(len = dims(1)), allocatable :: strings(:)

all tests pass.

I have attached a patch for completeness.