compile error for i686 since 3.6.5

[cheese1]

i get an compile-error when compiling for i686:

g++ -o generate_loop_macros generate_loop_macros.cpp -I.. -Wall -Wextra -Wfatal-errors -Werror=unknown-pragmas -Werror=unused-label -Wshadow -Wno-int-in-bool-context -Wno-bool-compare -std=c++11 -pedantic -Dcimg_use_vt100 -Dcimg_display=1   -lm -lX11 -lpthread 
In file included from curve_editor2d.cpp:46:
../CImg.h: In static member function ‘static bool cimg_library::cimg::type<_Float16>::is_inf(long double)’:
../CImg.h:3086:24: warning: ISO C++ does not allow converting to ‘_Float16’ from ‘const long double’ with greater conversion rank [-Wnarrowing]
 3086 |         return !is_nan(val) && (val<cimg::type<cimg_float16>::min() || val>cimg::type<cimg_float16>::max());
      |                        ^~~
../CImg.h:3086:23: error: invalid conversion to type ‘_Float16’ without option ‘-msse2’
 3086 |         return !is_nan(val) && (val<cimg::type<cimg_float16>::min() || val>cimg::type<cimg_float16>::max());
      |                 ~~~~~~^~~~~
compilation terminated due to -Wfatal-errors.
make[1]: *** [Makefile:323: curve_editor2d] Error 1
make[1]: *** Waiting for unfinished jobs....
In file included from dtmri_view3d.cpp:44:
../CImg.h: In static member function ‘static bool cimg_library::cimg::type<_Float16>::is_inf(long double)’:
../CImg.h:3086:24: warning: ISO C++ does not allow converting to ‘_Float16’ from ‘const long double’ with greater conversion rank [-Wnarrowing]
 3086 |         return !is_nan(val) && (val<cimg::type<cimg_float16>::min() || val>cimg::type<cimg_float16>::max());
      |                        ^~~
../CImg.h:3086:23: error: invalid conversion to type ‘_Float16’ without option ‘-msse2’
 3086 |         return !is_nan(val) && (val<cimg::type<cimg_float16>::min() || val>cimg::type<cimg_float16>::max());
      |                 ~~~~~~^~~~~

the complete compile-log can be found at:
https://kojipkgs.fedoraproject.org//work/tasks/7541/140027541/build.log

other arches compile fine:
https://koji.fedoraproject.org/koji/taskinfo?taskID=140027446

as far as i can see there was a change concerning float16 between 3.6.4 and 3.6.5

[dtschump]

Thanks for reporting.
Does this : 94f00eb
solve the issue ?

[cheese1]

compile works now:
https://kojipkgs.fedoraproject.org//work/tasks/9686/140119686/build.log

no further checks done

other archs compile, too:
https://koji.fedoraproject.org/koji/taskinfo?taskID=140119631

[cho-m]

Noting here that 94f00eb (thus latest http://cimg.eu/files/CImg_3.6.5.zip) is not compatible with imath >= 3.1.0 (AcademySoftwareFoundation/Imath@0ef9adc).

I was trying out change for Homebrew as we hit an issue with _Float16 on aarch64 Linux using GCC 12 (as _Float16 was only added in GCC 13).

However, using the new header to build G'MIC causes a failure:
CImg.h:62650:21: error: use of undeclared identifier 'cimg_float16'

Looks like some cimg_use_openexr-specific code assumes cimg_float16 is defined, but the check on _IMATH_H_ means it won't be defined when using latest OpenEXR/IMath.

[dtschump]

Thanks.
I've pushed 50674b2
to force defining cimg_float16 when openexr is enabled.
Does that work for you ?

[cheese1]

still compiles with the second patch.
i had to shorten the patch to only contain CImg.h, but not html/* as those are not there at compile-time.
and i have noticed that you republished a modified version 3.6.5, which might be bad, as 3.6.5 was already published. packagers might not detect the new version as being diffrent from the old one. but besides that: compiles.

[cho-m]

Does that work for you ?

Also works for Homebrew:

• gmic cimg 3.6.5 Homebrew/homebrew-core#258187