Re: mkoctfile on windows

[Kai Torben Ohlhus]

On 12/28/19 1:03 AM, JuanPi wrote:
> On Fri, Dec 27, 2019 at 2:41 PM JuanPi <address@hidden> wrote:
>>
>> Hi all,
>>
>> I noticed that compiling in GNU Octave 5.1.0 for windows seems to be 
>> possible.
>>
>> ## F77
>> First I found the problem that the environmental variable F77 (used in
>> some makefiles) is defined to f77 even if f77 doesn't exist, while
>> gfortran does exist, that is the lines
>>
>> F77 ?= gfortran
>>
>> would always have f77 even if f77 is not in the system. Forcing the
>> re-definition of F77 got some of my files compiled in windows
>>
>> Is this a bug?


In the autoconf manual [1] "F77" is regarded as "legacy Fortran 77
macros (F77), and modern Fortran macros (FC)".  In both my Linux
(openSUSE 15.1) and Win 10 VirtualBox (running Octave's cmdshell.bat to
get "make") the implicit variables "F77" and "FC" [2] contain the
default value "f77" even if it does not exist.  Thus not a bug I guess.


>>
>> ## mkoctfile failure
>> Most fo the files compile, but some of them (that compile in linux), I
>> get errors like th eones in the attached file.
>>


Your list is very long and several bug reports might come out, I guess.
 Is the use case of including "matrix.h" (described by you later) a
realistic scenario and worth the effort?


>> any idea what is causing this?
>>
>> Thanks
>> --
>> JuanPi Carbajal
>> https://goo.gl/ayiJzi
>>
>> -----
>> “An article about computational result is advertising, not
>> scholarship. The actual scholarship is the full software environment,
>> code and data, that produced  the  result.”
>> - Buckheit and Donoho
> 
>
> Dear all,
> 
> I have solved the problem.
> 
> 1. use mkoctfile instead of any specific compiler
> 2. when linking a mex file to objects generated from fortran source
> code, use -lgfortran when


This I think is a regression [3], as the Octave Fortran examples do not
compile either cleanly, as described in the manual [4] for 5.1.0 on Win
10 VirtualBox.  On Linux this does not seem to be required.


> 3. There ws a header file "matrix.h" included whic was causing all the
> troubles, I removed this include.
> 
> If you are using Octave on windows you can double chec and let me know
> if this package installs well from its repository. Run
> 
> pkg install 
> https://gitlab.com/hnickisch/gpml-matlab/uploads/e8bafa6308e81a1e61b8da51fef1a95a/gpml-4.2.0.tar.gz
> 
> there will be some warnings regarding missing docstrings but the
> package should be installed without issues. I tested on Windows 10
> using the installer (admin rights) and also using the zip file (after
> running post-install.bat and octave-firsttime.vbs).
> 


Installs cleanly on a Win 10 VirtualBox

>> pkg install
https://gitlab.com/hnickisch/gpml-matlab/uploads/e8bafa6308e81a1e61b8da51fef1a95a/gpml-4.2.0.tar.gz
warning: doc_cache_create: unusable help text found in file
'gpml_demoRegression'
warning: doc_cache_create: unusable help text found in file 'isLegal'
warning: doc_cache_create: unusable help text found in file 'precondDiag'
warning: doc_cache_create: unusable help text found in file 'precondTriu'
warning: doc_cache_create: unusable help text found in file
'precondTriuDiag'
For information about changes from previous versions of the gpml
package, run 'news gpml'.


> Regards,
> 
> Thanks
> 
> 



[1]
https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/autoconf.html#Fortran-Compiler
[2]
https://www.gnu.org/software/make/manual/html_node/Implicit-Variables.html
[3] https://savannah.gnu.org/bugs/?49548
[4]
https://octave.org/doc/v5.1.0/Calling-External-Code-from-Oct_002dFiles.html