[Tgcmgroup] Having a little trouble compiling

[Erik Wilson]

Hi Lorenzo,

The unlink and system functions are intrinsic to gfortran, but are 
marked as external in util.F.  If you remove or comment the 
"integer,external :: system" directive on line 396, and 
"integer,external :: unlink" directive on line 433 of util.F, that 
should allow those functions to be used properly.

Cheers,
  -Erik

On 01/14/2011 05:17 PM, Lorenzo Luengo wrote:
> Thanks Ben and Frederick,
>
> I know what the system function tries to do, i tried fixing it linking
> against libgfortran, that supplies "system" and "unlink" functions, but
> the linking step of the model is looking for the "system_" and "unlink_"
> functions. Also tried some compiler switches like "-fno-underscore", but
> when i compile using this switch, it breaks compatibility with netcdf (a
> HUGE drawback), that uses that style of underscored function names in
> "libnetcdff.a".
>
> Next thing that I will try will be compiling my own version of
> libgfortran (a long recompilation of gcc), to build the library using
> the underscores. Maybe ubuntu defaults for libgfortran uses
> "-fno-underscores" switch and that is causing this trouble.
>
> The only modifications I've been making are in the
> "tiegcm1.93-linux.job" and tiegcm1.93/scripts/Makefile" tweaking some
> flags and paths to the libraries. Most of the build process has gone
> smoothly.
>
> I'll let you know if i can make it work.
>
>
> Thanks for all you advice.
>
> PD: Did both of you reply directly to me? I think is better replying to
> the whole list, for pipermail to archive this.
>
> El 14-01-2011 14:49, Ben Foster escribió:
>> Lorenzo,
>>
>> The "system" function attempts to escape to the OS and execute
>> a shell command. It may be platform dependent (different names),
>> so the model checks the current OS to make the correct call. These
>> calls are mostly for file searching and manipulation on the local
>> disk e.g., the "unlink" function it complained about.
>>
>> The OS is obtained by the Makefile with the "uname -s" command.
>> If you can get the results of this command, or its equivalent
>> in Ubuntu, we can add it to the list of OS's known to the model.
>>
>> When you get the OS, look in src/util.F and search on "function isystem"
>> This is the local function that calls system according to the OS,e.g.,
>> you see UNICOS, AIX, OSF1, SUN, LINUX, IRIX, etc. Here you can
>> add your Ubuntu or UBUNTU, or whatever. Same for iunlink. Several
>> functions in util.F
>> check the OS with #ifdef statements for various reasons. Actually,
>> you could simply add or change the default case in the #ifdef statements
>> in util.F.
>>
>> Having said this, most of the system and unlink calls refer to
>> local NCAR file archiving, mss, searching the disk, etc., and do not
>> really belong in the Community model. I am planning on removing it
>> soon, and rely only on simple fortran intrinsic file i/o. However, there
>> are other functions in util.F that depend on the OS, so you will probably
>> need to add the new OS, or defaults, to the conditionals in util.F.
>>
>> Let me know how it goes.
>>
>> --Ben
>>
>>
>> On 1/13/2011 9:44 PM, Lorenzo Luengo C. wrote:
>>> Hi everyone
>>>
>>> Today i tried to compile the model in a VM running Ubuntu, using
>>> gfortran.
>>>
>>> After installing some required libraries (mpich2, netcdf, zlib), and
>>> tweaking some compiler flags, it compiles, but it's not linking.
>>>
>>> The error is an undefined reference to functions named system_ and
>>> unlink_ (notice the underscores). I couldn't find a suitable library
>>> providing this functions, I tried using libf2c, but it did not work.
>>>
>>> Any advice on how to filfull this requirement?
>>>
>>> Thanks
>>>
>