On 2015-09-27 16:47, T. Modes wrote:
>
>
> Am Sonntag, 27. September 2015 15:13:19 UTC+2 schrieb jdx:
>
> > * The changes break the installation of wxWidgets in MSVC version.
> You
> > did probably test with a existing build and not from a clean one.
> Yes, I use "blessed" version for MSVC++ 2010 from
>
http://sourceforge.net/projects/wxwindows/files/3.0.2/binaries/
> <
http://sourceforge.net/projects/wxwindows/files/3.0.2/binaries/>. They
> have vc100 suffix.
>
>
> But your code will then only work with the 32 bit version, the 64 bit
> version does not work with your code.
> Until now we have used a self compiled wxWidgets library, and for this
> use case the filename work as there.
OK, this needs more work. But does the original code work properly with
32- and 64-bit libraries? Because suffix "custom" means nothing.
> > * Usage of empty ${wxWidgets_CONFIG_OPTIONS} variable.
> Where? Anyway, variable wxWidgets_CONFIG_OPTIONS contains options which
> are passed to wx-config utility, it may be empty and by default it is
> empty. See FindwxWidgets.cmake for more information.
>
>
> But we don't set it currently. So it remains empty and adding it makes
> all more complicated than it is already. So we can currently ignore it.
For example I set this variable at CMake's command line. It is useful if
you have multiple wxWidgets configurations (eg. "debug" and "release")
or if your actual wxWidgets prefix is other that the one specified at
configure stage.
> > * Install target for lapack is also using gcc files also for MSVC:
> >
> > IF(NOT MINGW) INSTALL(FILES ${LIBGCC_DLL}...
> Yes, because I use LAPACK built by myself which dynamically links to
> libgcc, libgfortran, etc. just like the "official" one:
>
http://icl.cs.utk.edu/lapack-for-windows/lapack/
> <
http://icl.cs.utk.edu/lapack-for-windows/lapack/>. However IMO it's
> not a
> big deal - I am quite sure that it is possible to build LAPACK's DLLs
> with libgcc and libgfortran statically linked.
>
>
> But even when using the prebuilt version you can't pull in files from
> gcc (because it is not installed when using MSVC compiler, there are
> then only the dll).
In fact gcc is installed. :-) But I get your point. The proper way is to
distribute libgfortran.dll & Co. with LAPACK's DLLs or link it
statically into them.
> LAPACK can be ignored, the code which is improved when compiling with
> LAPACK, is not used by Hugin.
> So these dependencies can be ignored. Maybe we should remove the LAPACK
> stuff from Hugins Cmake system.
Hmmm... I build Hugin with -DENABLE_LAPACK=ON and Dependency Walker
shows me that libhuginbase.dll depends on liblapack.dll.
> No. In official jpeg-9a, file jmorecfg.h line 245 reads:
> /* a function referenced thru EXTERNs: */
> #define GLOBAL(type) type
>
> So your are using a patched version.
Indeed I am. :-) For some reason I added to jmorecfg.h usual
__declspec(dllexport/dllimport) stuff.
> With MSVC libtiff_i.lib is the dynamic version and libtiff.lib the
> static one. When searching for both names in the same line, this can
> pull in the static version in the dynamic build scenario (as already
> stated in the comment directly above). This I want avoid and can
> therefore not search for both in your way. I committed a slightly
> modified version.
Oh, there is my fault to some extent because I have libtiff with
non-standard file name. By default, when built with MinGW, libtiff's
4.0.4 DLL is called libtiff-5.dll (five, not four) and I have changed it
to simple libtiff.dll. Static library is called libtiff.a and import
library is libtiff.dll.a.
Three new patches in the attachment. CMakeLists.diff removes unnecessary
wxWidgets related definitions at src directory level and two other
patches are critical - without them compilation and/or linking fails due
to some symbols redefinition.
Jan