Re: Interfacing Ada with C

[Simon Wright <simon@pushface.org>]

Ada novice <posts@gmx.us> writes:

> Success here with the testing! Thanks for all your very kind help
> Simon. First, I would like to draw your attention on 1 point. At
> first, I was copying your raw files into AdaGIDE in order to save
> them. Then running gnatmake -p -P test_extensions did produce a lots
> of error messages: incorrect line terminator. I remember getting this
> message before and what I did was to save the files in GPS and the
> error messages disappear. I think in AdaGIDE pasting raw text can
> cause some incompatibility due to the file system used. This is why I
> guess I got this error: incorrect line terminator. I can't explain
> this but I don't have it in GPS.

Was this perhaps with the ada-numerics* files? there may be some -gnatpg
problem with inconsistent line terminators, I guess GPS is clever enough
to know that it's going to be anuisance and fix it.

> After successfully saving all your raw files within GPS, I ran
> gnatmake -p -P test_extensions without any warnings/error messages and
> the following were created:
>
> A.) A lib folder is created at the same hierarchy level as src and
> test. It has 2 files:
> 1.libada_math.dll
> 2.ada-numerics-generic_complex_arrays-extensions.ali
>
> B.) In the src folder a .build folder is created with 2 files:
> 1. ada-numerics-generic_complex_arrays-extensions.ali
> 2. ada-numerics-generic_complex_arrays-extensions.o
>
> C.) In the test folder a .build folder is created with 6 files:
> 1. b~test_extensions.adb
> 2. b~test_extensions.ads
> 3. b~test_extensions.ali
> 4. b~test_extensions.o
> 5. test_extensions.ali
> 6. test_extensions.o
>
> Then I was able to build using Build main: test_extensions.adb
> successfully.

This looks as expected. NB that's a sort of half-baked library; to make
a proper dll * include files it needs to be a 'stand-alone library' in
GNAT terms. From the user's point of view it Just Works (I did have a
problem on a Debian machine with GCC 4.3.2, didn't automatically find
the shared library). Anyway, if it works for you on windows don't worry
about it!

> Now for the outputs I get:
>
>  2.00000E+00  4.00000E+00
>  2.00000E+00 -4.00000E+00
>  9.99999E-01  1.10569E-07
>
> -6.00000E+00
> -4.00000E+00
> -2.00000E+00
> -1.21279E-07
>  2.00000E+00
>  4.00000E+00
>  6.00000E+00
>
>  6.00000E+00  0.00000E+00
>  4.00000E+00  0.00000E+00
> -6.00000E+00  0.00000E+00
>  2.00000E+00  0.00000E+00
> -3.17504E-07  0.00000E+00
> -4.00000E+00  0.00000E+00
> -2.00000E+00  0.00000E+00
>
> The second set of results are for the Test16 which are ok. What about
> the third set of results? What are these?

The second set are with the package that comes with the compiler, that
takes a symmetric complex matrix (which this is - needs to be complex to
check the new code) and outputs a real eigenvalue vector.

The third set are with the new package, that takes a general complex
matrix and outputs a complex eigenvalue vector.

> I didn't add -gnatpg for the compiler switch. I had these: -gnatqQ -
> gnat05 and I can compile fine and output the results. Is the -gnatpg a
> necessity?

Yes *but only in the src/ directory*, which is why it's in
ada_math_build.gpr & not in test_extensions.gpr.

This is part of a clean compile...

   gcc -c -g -fPIC -gnatpg -gnatqQ -gnat05 -I- -gnatA /Users/simon/gnat-math-extn/src/ada-numerics-generic_complex_arrays-extensions.adb
   gcc -c -g -gnatqQ -gnat05 -I- -gnatA /Users/simon/gnat-math-extn/test/test_extensions.adb

and the first line, for the library, has the additional flags -fPIC
(because, on this architecture, you need position-independent code for a
shared library) and -gnatpg (from ada_math_build.gpr, as discussed).

> Thanks very much for helping me with the compilation. Now it's good
> that some more testing is done on different cases. Do you plan to add
> the output for eigenvectors also?

Seemed like a plan! Also real non-symmetric, I think.

I found the LAPACK test suite described here -
http://www.netlib.org/lapack/lawns/lawn41.ps - could be useful though it
could also be a pain to translate.