Problem compiling NONMEM with mingw gfortran 4.3.0 builds

[Steve Chapel]

Tobias Burnus wrote:
> Hi,
>
> Steve Chapel wrote:
>   
>> I'd actually rather use a stable build of gfortran than current
>> development builds. Is there any chance of someone providing mingw
>> binaries of gfortran 4.0.3
>>     
> Definitely not for 4.0.x; gfortran was first included in 4.0 and all
> 4.0.x versions were still rough and buggy. Additionally, GCC does not
> maintain 4.0.x anymore.
>   
>> or gfortran 4.1.2?
>>     
> That would work, but I would recommend to use 4.2.x, which has many
> Fortran improvements, is a stable version and already used as system
> compiler of some Linux distributions.
>
> As I don't have a working Windows, someone else has to build gfortran
> 4.2.x/Mingw ... FX?
>
> Nonetheless, the developer version gfortran 4.3.x usually works and has
> only rarely worse bugs than 4.2.x. (There are some/sometimes
> regressions, though.)
>   
In many cases, I could use a developer version of software. In my 
particular case, however, we need to comply with some government 
regulations regarding software use. We can be audited, and if during the 
audit it came out that we were relying on a developer version of a 
compiler, that wouldn't look too good, as there is very little 
verification that it is producing the correct results. I'm not sure if 
we can even use a compiler that we built ourselves from official source 
code, especially if no one else uses it. Right now, I'm simply 
evaluating whether gfortran will work and what performance increase (if 
any) we would get if we decide to buy the Intel Fortran compiler. A 
developer version will work for evaluation purposes, but not for 
production runs.

We could use official or semi-official builds of stable releases for 
production, especially if those releases were well-known and used by 
many users. I actually tried DJGPP builds of gfortran first, but ran 
into the 127 character command line limit. One of my problems with MinGW 
builds is that there are so few pre-built binaries for Windows, and they 
don't seem to be very well known. I think the MinGW could benefit users 
by providing pre-built binaries as DJGPP does, and those users could in 
turn help MinGW by providing more extensive testing of the releases.

-- Steve