[CMake] help with shared-lib make failure

[Andreas Naumann]

Am 18.06.2012 08:34, schrieb luxInteg:
> On Sunday 17 June 2012 14:13:00 Andreas Naumann wrote:
>
>    
>>> #----------build2: gs_test/results.c.  build  disabled   near beginning
>>> but comes in ~65%   of build
>>> [ 65%] Building C object Lib/CMakeFiles/gsl.dir/__/test/results.c.o
>>> cd /home/TESTS/BUILDgsl/Lib&&   /usr/bin/gcc  -Dgsl_EXPORTS
>>> -DHAVE_CONFIG_H - DPIC -m64 -fPIC  -msse2 -fPIC -I~SRS/blas -I~SRS/block
>>> -I~SRS/bspline - I~SRS/cblas -I~SRS/cdf -I~SRS/cheb -I~SRS/deriv
>>> -I~SRS/dht -I~SRS/eigen - I~SRS/err -I~SRS/fit -I~SRS/gsl
>>> -I~SRS/histogram -I~SRS/iee-utils - I~SRS/integration
>>> -I~SRS/interpolation -I~SRS/linalg -I~SRS/matrix -I~SRS/min -I~SRS/monte
>>> -I~SRS/multifit -I~SRS/multimin -I~SRS/multiroots -I~SRS/multiset
>>> -I~SRS/ntuple -I~SRS/ode-initval -I~SRS/permuation -I~SRS/poly
>>> -I~SRS/qrng - I~SRS/randist -I~SRS/rng -I~SRS/roots -I~SRS/siman
>>> -I~SRS/sort -I~SRS/specfunc -I~SRS/statistics -I~SRS/sum -I~SRS/sys
>>> -I~SRS/test -I~SRS/utils -I~SRS/vector -I~SRS/wavelet
>>> -I/home/TESTS/BUILDgsl -I~SRS   -msse2 -mfpmath=sse -msse3 - mfpmath=sse
>>> -Wall -o CMakeFiles/gsl.dir/__/test/results.c.o   -c ~SRS/test/results.c
>>> ~SRS/test/results.c: In function 'gsl_test':
>>> ~SRS/test/results.c:89: error: 'va_list' undeclared (first use in this
>>> function)
>>> ~SRS/test/results.c:89: error: (Each undeclared identifier is reported
>>> only once
>>> ~SRS/test/results.c:89: error: for each function it appears in.)
>>> ~SRS/test/results.c:89: error: expected ';' before 'ap'
>>> ~SRS/test/results.c:93: warning: implicit declaration of function
>>> 'va_start' ~SRS/test/results.c:93: error: 'ap' undeclared (first use in
>>> this function) ~SRS/test/results.c:96: warning: implicit declaration of
>>> function 'va_end' ~SRS/test/results.c: In function 'gsl_test_rel':
>>> ~SRS/test/results.c:149: error: 'va_list' undeclared (first use in this
>>> function)
>>> ~SRS/test/results.c:149: error: expected ';' before 'ap'
>>> ~SRS/test/results.c:153: error: 'ap' undeclared (first use in this
>>> function) ~SRS/test/results.c: In function 'gsl_test_abs':
>>> ~SRS/test/results.c:225: error: 'va_list' undeclared (first use in this
>>> function)
>>> ~SRS/test/results.c:225: error: expected ';' before 'ap'
>>> ~SRS/test/results.c:230: error: 'ap' undeclared (first use in this
>>> function) ~SRS/test/results.c: In function 'gsl_test_factor':
>>> ~SRS/test/results.c:302: error: 'va_list' undeclared (first use in this
>>> function)
>>> ~SRS/test/results.c:302: error: expected ';' before 'ap'
>>> ~SRS/test/results.c:307: error: 'ap' undeclared (first use in this
>>> function) ~SRS/test/results.c: In function 'gsl_test_int':
>>> ~SRS/test/results.c:357: error: 'va_list' undeclared (first use in this
>>> function)
>>> ~SRS/test/results.c:357: error: expected ';' before 'ap'
>>> ~SRS/test/results.c:362: error: 'ap' undeclared (first use in this
>>> function) ~SRS/test/results.c: In function 'gsl_test_str':
>>> ~SRS/test/results.c:401: error: 'va_list' undeclared (first use in this
>>> function)
>>> ~SRS/test/results.c:401: error: expected ';' before 'ap'
>>> ~SRS/test/results.c:406: error: 'ap' undeclared (first use in this
>>> function) make[2]: *** [Lib/CMakeFiles/gsl.dir/__/test/results.c.o]
>>> Error 1 make[2]: Leaving directory `/home/TESTS/BUILDgsl'
>>> make[1]: *** [Lib/CMakeFiles/gsl.dir/all] Error 2
>>> make[1]: Leaving directory `/home/TESTS/BUILDgsl'
>>> make: *** [all] Error 2
>>>
>>>
>>>
>>> suggestions welcome
>>>
>>> sincerely
>>> luxInteg
>>>        
>> The comp misses the declaration of the type "va_list". if those errors
>> are the first ones, then this type is simply never declared. If it works
>> with the static library, you should have look which preprocessor flags
>> are used for the static libraries and compare them with the shared ones.
>>
>>      
> The  preprocessor flag appears to be -M.  When this is sumstituted  a 'patched
> version of results.c does indeed compile and the build progresses to  ~70%
> where   th results.c is again compiled  with a load of other files  into a
> large library and there   are no obvius ways   to isolate its  compilation to
> add the -M definition.  I tried set_source_file_properties(PROPERTIES compile
> definitions "-M")  but this adds a -D-M  which does not seem to be the same -
> M.
>
> Anyway the problem seems to be a troublesome test/results.c file.
> (It is attached for your info )
>    the unpatched version gives this response at the start of the build :-
>
> #----------------
> In file included from ~/TESTS/gsl-1.15/test/results.c:32:
> /usr/lib/gcc/x86_64-unknown-linux-gnu/4.4.2/include/varargs.h:4:2: error:
> #error "GCC no longer implements<varargs.h>."
> /usr/lib/gcc/x86_64-unknown-linux-gnu/4.4.2/include/varargs.h:5:2: error:
> #error "Revise your code to use<stdarg.h>."
> make[2]: *** [test/CMakeFiles/gsltest-static.dir/results.c.o] Error 1
> #------------
>
> the file seems to need  updating   (extensively patched ) for newer compilers
> so any suggestiions  you have  will be gratefully received.
>
>
> sincerely
> luxInteg
>    
I think it is a bad idea to rely on the -M flag, because cmake tries to 
figure out the dependencies..

You could start with defining STDC_HEADERS, or let cmake look for the 
needed headers and set the flag appropriatly.

I start to send the conversation to the list again. Possibly some others 
had similiar trouble when switching from the autotools suite to cmake.

Andreas