Compiling under Windows: Difference between revisions
m (→Compiling EQC) |
|||
Line 173: | Line 173: | ||
* Run <code>./configure CC=cl CXX=cl LIB=/vc/lib:/sdk/lib CPPFLAGS="-DEBUG -MD -EHsc -I/vc/include -I/sdk/include" PKG_CONFIG_PATH=/usr/local/lib/pkgconfig</code> | * Run <code>./configure CC=cl CXX=cl LIB=/vc/lib:/sdk/lib CPPFLAGS="-DEBUG -MD -EHsc -I/vc/include -I/sdk/include" PKG_CONFIG_PATH=/usr/local/lib/pkgconfig</code> | ||
* Run LIB=/vc/lib:/sdk/lib make | * Run LIB=/vc/lib:/sdk/lib make | ||
* | * Ignore the failure to compile lt-eqc.c and use .libs/eqc.exe instead |
Revision as of 09:09, 23 December 2010
Preparing the Openoffice SDK with Microsoft Visual C++
To build iMath for Windows, you need the Windows Openoffice SDK.
- Download and install the Openoffice SDK
- The following is explained in detail in "SDK installation directory"\sdk\docs\install.html
- Install Microsoft Visual C++ (the free express version will do, then you also need to install the Microsoft Visual Studio 2008 Redistributable Package
- Install make (e.g. copy C:\MinGW\bin\mingw32-make.exe to WINDOWS\system32 and rename it to make.exe)
- Install info-zip (e.g. download from [1] and put zip.exe into C:\Windows\system32)
- To test the SDK installation, open a command prompt and cd to the ...\sdk\examples\cpp\DocumentLoader directory.
- Run ...\sdk\setsdkenv_windows.bat to configure your environment. This should create a custom batch file somewhere in your home directory (usually Application Data\Openoffice\...). Give it the correct paths to the make and zip executables
- I had to edit thecustom batch file and add a correct URE path:
set OO_SDK_URE_HOME=C:\Programme\OpenOffice.org 3\URE
- Now try to build the example by typing make, if this is not successful fix the problems before continuing.
- The next time you want to use the SDK, just run the custom batch file to set the environment variables required
Preparing MSYS and Mingw
- Get the mingw installer from here
- Use the pre-packaged lists for the first install
- Do mingw-get update and mingw-get install msys-base mingw32-base. This should update libtool to 2.4
- Click on the MSYS icon to start a Bourne shell
- Add .../Microsoft Visual Studio 9.0/Common7/IDE and .../Microsoft Visual Studio 9.0/VC/BIN to PATH (e.g
PATH=$PATH:"/c/Programme/Microsoft Visual Studio 9.0/Common7/IDE":"/c/Programme/Microsoft Visual Studio 9.0/VC/bin"
). You can find these paths in vsvars32.bat somewhere in the Visual Studio installation directory. - If you want to use cl (the Microsoft compiler) to compile something, copy .../Microsoft Visual Studio 9.0/include and lib to msys vc/include and vc/lib (or find a better way to make CPPFLAGS accept path names with spaces)
- Copy Microsoft SDKs/Windows/Vxxx/Include and Lib to msys /sdk/include and /sdk/Lib (or find a better way to set the cl.exe LIB path)
Compiling gmp with gcc
Download and unpack gmp into your msys home directory.
- Change into the new directory.
- For a shared library build, run configure with
--disable-static --enable-shared
. Then do make, make check and make install - Change into the .libs subdirectory
- Make sure the MSVC binaries are in your path and run
lib /def:libgmp-3.dll.def /out:libgmp-3.lib
- Copy libgmp-3.exp, libgmp-3.dll.def and libgmp-3.lib into /usr/local/lib.
- Do make distclean
- For a static library build, run configure with
--enable-static --disable- shared
(note that it is not possible to build both static and shared libraries in one compilation run). - Do make, make check and make install
- Note that make install overwrites the libgmp.la for dynamic linking in /usr/local/lib.
- Download objconv objconv and put the executable somewhere into your msys path.
- Change into the .libs subdirectory
- Run
objconv -fcoff -nr:___chkstk:__chkstk libgmp.a libgmp.lib
to make the symbols compatible with MSVC - Copy the new libgmp.lib into /usr/local/lib
To compile a test program with MSVC
- Open a DOS command shell and go to the gmp directory.
- You can find some example files in the demos subdirectory of gmp.
- For a dynamically linked program, run
cl /MD -I \vc\include -I \sdk\include -I \local\include isprime.c /link /LIBPATH: \vc\lib /LIBPATH: \sdk\lib LIBPATH: \local\lib libgmp-3.lib
- For a static link, substitute libgmp-3.lib with libgmp.lib
Compilation with Microsoft Visual C++
This is what I tried to get a native MSVC compile (because Openoffice is compiled with MSVC), in spite of being warned that CLN is not compileable with anything except gcc...
Preparations
- Install mingw (the minimal version will do since we will be using the Microsoft compiler)
- Set your PATH as described above in the gcc compile section and create the msys /vc and /sdk directories
- Create the /vc/include/sys directory and put /vc/include/time.h into it
Compiling CLN
Download the CLN source code and unpack it into your msys home directory
- Change to the source directory
- run
./configure CC=cl CXX=cl CCAS=cl CPPFLAGS="-DEBUG -DNO_ASM -EHsc -MD -I/vc/include -I/sdk/include" LIB=/vc/lib:/sdk/lib
- make will stop and complain about missing files cl_asm.obj and cl_asm_GF2.obj
- copying cl_asm.o from a gcc compile to cl_asm.obj gives lots of linker errors, don't try that
- change to the src subdirectory
- Remove cl_asm.lo and cl_asm_GF2.lo
- Edit the Makefile and find the lines starting with
cl_asm.lo:
andcl_asm_GF2.lo:
. Remove the comment from the first line below them - Change cl_asm.S and cl_asm_GF2.S to cl_asm.cc and cl_asm_GF2.cc in the first lines and in the lines you just uncommented (because MSVC doesn't recognize the .S extension). Total of five changes for each.
- cp base/digitseq/cl_asm.S to base/digitseq/cl_asm.cc. cl_asm_GF2.cc already exists in my distribution and is identical to cl_asm_GF2.S
- run make cl_asm.lo and make cl_asm_GF2.lo. This should create cl_asm.obj and cl_asm_GF2.obj
- If you get a libtool error about not finding cygpath, edit libtool in the cln root directory and find the line reading
fix_srcfile_path="\`cygpath -w \"\$srcfile\"\`
. Change it tofix_srcfile_path="$srcfile"
.
- Run make again
The build will finish with some linker warnings about duplicate definitions and create the library in .libs. The linker errors are (hopefully) all harmless and have to do with different handling of inlined functions by gcc and MSVC
- bool zerop(const cl_RA&): As soon as more than one file includes cl_RA.h, inline bool zerop(..) is defined twice and produces a linker warning
- cl_I numerator (const cl_RA&): same as above
- bool CL_FLATTEN minusp (const cl_RA&): same as above
- const cl_I denominator (const cl_RA&): same as above
- bool zerop (const cl_I&): As soon as more than one file includes cl_I.h, inline bool zerop(..) is defined twice and produces a linker warning
- bool minusp (const cl_I&): Same as zerop
- const cl_LF operator+ (const cl_LF&, const cl_LF&): Defined in cl_LF_2plus.cc and as an inline version in cl_LF.h, these clash when cl_R_mul.cc includes cl_LF.h
- same for operator-
When you try to use the library created, there will be lots of unresolved externals when linking. The reason is that libtool actually does two links when creating the library (why?) and the second link has less files in its list (why?). Therefore, go to the Makefile again and edit the following entry (the second and following lines must start with a tab character):
libcln.la: $(libcln_la_OBJECTS) $(libcln_la_DEPENDENCIES) $(libcln_la_LINK) -rpath $(libdir) $(libcln_la_OBJECTS) $(libcln_la_LIBADD) $(LIBS) list='$(libcln_la_OBJECTS)'; \ libobjects="`echo $$list | sed 's/\\.lo/.obj/g'`"; \ lib -OUT:.libs/libcln.lib $$libobjects $(libcln_la_LIBADD) $(LIBS)
Remove src/.libs/libcln.lib, run touch cl_asm.lo (or any other dependency of the library) and run make again.
Now try compiling the examples:
- Change to the examples subdirectory
- Edit the Makefile and change the entry for CXXLINK by adding
/vc/lib/msvcrt.lib /vc/lib/msvcprt.lib /vc/lib/oldnames.lib /sdk/lib/Kernel32.lib /sdk/lib/Uuid.lib
at the end. If you can figure out how to give the library directories to configure please let me know! - Also, sometimes the compiler flag
-MD
given to configure seems to get lost in the examples Makefile. Add it to CXXCOMPILE if you get complaints about a missing libcpmt.lib
The linker will fail with some unresolved externals. The reason is the different handling of extern declarations by gcc and MSVC. The errors can be removed by editing the source:
libcln.lib(cl_I_from_digits.obj) : error LNK2019: unresolved external symbol _mulu32_high referenced in function "class cln::cl_I const __cdecl cln::digits_to_I(char const *,unsigned int,unsigned long)" (?digits_to_I@cln@@YA?BVcl_I@1@PBDIK@Z)
.- cl_low.h defines
extern "C" uint32 mulu32_high;
- Solution: Move this declaration outside of the namespace cln in cl_low.h (e.g. to the beginning of the file)
- cl_low.h defines
- The same applies for
_divu_32_rest
and_divu_16_rest
libcln.lib(cl_F_dprint.obj) : error LNK2019: unresolved external symbol "struct cln::cl_anything * __cdecl cl_I_constructor_from_UL(unsigned long)" (?cl_I_constructor_from_UL@@YAPAUcl_anything@cln@@K@Z) referenced in function "public: __thiscall cln::cl_I::cl_I(unsigned long)" (??0cl_I@cln@@QAE@K@Z)
.- cl_I.h declares
extern cl_private_thing cl_I_constructor_from_UL (uint32 wert);
andextern cl_private_thing cl_I_constructor_from_Q (sint64 wert);
somewhere in the file - Comment out these declarations
- Replace them by the following at the beginning of the file:
- cl_I.h declares
extern cln::cl_private_thing cl_I_constructor_from_UL (uint32 wert); extern cln::cl_private_thing cl_I_constructor_from_Q (sint64 wert);
- Edit cl_I_from_UL.cc and comment out the namespace declaration
namespace cln {
and} // namespace cln
- Add
using namespace cln;
- Do the same for cl_I_from_Q.cc
- Edit cl_I_from_UL.cc and comment out the namespace declaration
Now pi.exe compiles and links and calculates PI (whether correctly or not, I did not check). If it doesn't, you might have to remove the .lo files which the linker complains about and re-make in the src subdirectory so that the source code modifications find their way into the library.
The other examples need more modifications of the source along the same lines to link. Only e.cc presents new problems
- Add
#include "../src/float/lfloat/cl_LF.h"
to the beginning of the file - Comment out the
extern cl_LF cl_I_to_LF(const cl_I&, uintC);
declaration lower down - Change the call to
log(10)
tolog ((double)10))
- Edit cl_LF.h, comment out
extern const cl_LF cl_I_to_LF (const cl_I& x, uintC len);
and addextern const cln::cl_LF cl_I_to_LF (const cln::cl_I& x, uintC len);
to the beginning of the file, outside the namespace cln - Edit cl_LF_from_I.cc, remove
namespace cln{}
and addusing namespace cln;
instead
To ensure that all the files depending on cl_LF.h etc. get recompiled, it is simplest (but takes a while) to do a make clean and then a new make.
Try compiling the checks:
- Edit the makefile in the tests subdirectory and manually add the libraries as above
- Run "make check"
- On my compile, this compiled without a flaw and said "All two tests passed"
If you want to use the gmp library as compiled and modified by objconv above, add --with-gmp=/usr/local LDFLAGS=/usr/local/lib/libgmp.lib
to the configure invocation. You might have to create a copy of libgmp.lib named libgmp.a in /usr/local/lib because make will link to both of them.
To get a successful GiNaC compile, more edits of the source code for missing externals are necessary
- Edit include/cln/integer_io.h
- Find four lines with
extern print_integer
and move them to the top of the file, outside of namespace cln - Find the line with
extern cl_print_flags default_print_flags
and move it to the top of the file - Add
cln::
to all occurences ofcl_print_flags
andcl_I
which you moved out of the namespace
- Find four lines with
- Edit include/cln/output.h
- Move the line
extern cl_print_flags default_print_flags;
to the top of the file, outside of namespace cln - Add
cln::
beforecl_print_flags
- Move the line
- Do the same for float_io.h, lfloat_io.h, dfloat_io.h, sfloat_io.h, ffloat_io.h, complex_io.h, real_io.h and rational_io.h
- Edit cl_I_[a-d]print.cc, remove the
namespace cln
and addusing namespace cln;
to the top of the files - Do the same for cl_I_print.cc
- Edit cl_prin_globals.cc and move
cl_print_flags default_print_flags;
outside of namespace cln. Addcln::
beforecl_print_flags
- Do
make clean; make
to catch all the dependencies
Actually, there are lots more header files which would need similar edits, but I have listed only those required to get GiNaC to build.
Compiling GiNaC
Download the Ginac source code from git and unpack it into your msys home directory
- Download pkg-config, gettext and glib2 from the GTK project. Unzip the files and install them under /MinGW/bin
- Change to the source directory
- Run
./configure CC=cl CXX=cl CPPFLAGS="-DEBUG -MD -EHsc -I/vc/include -I/sdk/include" LIB=/vc/lib:/sdk/lib PKG_CONFIG_PATH=/usr/local/lib/pkgconfig --disable-shared --enable-static
- If configure chokes, find pkg.m4 on the web and put it in MinGW/share/aclocal
- If you have not installed libreadline, then configure will give a warning. You can ignore this if you do not plan to build the ginsh.
- run make
- The linker complains about "no public symbols existing" in registrar.obj, which seems to be OK since the source file has no code.
- Make will fail in the ginsh subdirectory if you have not installed libreadline. Run make -t in this directory to ignore the problem.
- Running
LIB=/vc/lib:/sdk/lib make check
passes all 58 tests successfully
Compiling EQC
Download the EQC source code and unpack it into your msys home directory
- Change to the eqc directory
- Edit src/Makefile.am and comment out the line
CXX := /usr/bin/ccache $(CXX)
if you don't have ccache installed - Run
autoreconf -i
to get the newest libtool files - Run
./configure CC=cl CXX=cl LIB=/vc/lib:/sdk/lib CPPFLAGS="-DEBUG -MD -EHsc -I/vc/include -I/sdk/include" PKG_CONFIG_PATH=/usr/local/lib/pkgconfig
- Run LIB=/vc/lib:/sdk/lib make
- Ignore the failure to compile lt-eqc.c and use .libs/eqc.exe instead