The 'main()' function in svpaint.cpp is overriding my main function

[namniav]

Environment

• Tesseract Version: 4.1.1, installed with vcpkg
• Commit Number: unknown
• Platform: Windows 10 21H2, 19044.1766, 64 bit

Current Behavior:

// main.cpp
#include <cstdio>
int main(int, const char*[]) {
  std::puts("hello tesseract");
  return 0;
}

$ g++ -L /k/vcpkg/installed/x64-windows/lib -ltesseract41 -lleptonica-1.82.0 main.cpp
K:/Qt6/Tools/mingw1120_64/bin/../lib/gcc/x86_64-w64-mingw32/11.2.0/../../../../x86_64-w64-mingw32/bin/ld.exe:
  C:\Users\namniav\AppData\Local\Temp\ccSM394z.o:main.cpp:(.text+0x0): multiple definition of `main';
  K:/vcpkg/installed/x64-windows/lib/tesseract41.lib(tesseract41.dll):(.text+0x0): first defined here
collect2.exe: error: ld returned 1 exit status

This example reports linkage error because tesseract41.dll is linked before main.o. But when it is linked to a Qt&CMake project, it is overriding my main function silently and writes Starting java -Xms512m -Xmx1024m -Djava.library.path=. -jar ./ScrollView.jar to log.

Expected Behavior:

I think shared library should not carry a main function.

[zdenop]

tesseract 4.x is the unsupported version. Please use the recent tesseract version (5.2).
I tried clang:

clang-cl main.cpp -o test.exe -I F:/win64/include F:/win64/lib/tesseract51.lib

and there was no problem.
I tried VS:

cl main.cpp -I F:/win64/include /link /LIBPATH:F:\win64\lib tesseract51.lib

and there is no problem.
I tried MSYS2 (with the default installation of tesseract 5.1):

@level MINGW64 /F/Projects/Community/tesseract
# g++ main.cpp -I /clang64/include/ -L /clang64/lib/ -l tesseract

Actually, tesseract executable (also 4.x version) has the main function and links to link tesseract library without a problem on any platform...

tesseract/src/api/tesseractmain.cpp

Line 599 in 82a7a87

int main(int argc, char** argv) {

[namniav]

@zdenop
Currently 4.1.1 is the only version supported by vcpkg. It is the easiest and stable way to build tesseract. If the issue was fixed in 5.x, I can try building a newer version.

But all your tries were linking tesseract AFTER main.o(or main.cpp), can you try linking tesseract FIRST? This is the default behavior when using CMake: when A depends B, B is linked before A itself otherwise there will be undefined references in A if it is referencing symbols from B. But MSVC's linking algorithm is different and order insensitive when linking shared libraries, if I recall correctly. Anyway, when linking/loading multiple shared libraries, symbols with same names from later libraries are ignored.

[zdenop]

This is the default behavior when using CMake:

Maybe I am missing something but our cmake builds works:

• vcpkg (cmake) build works also for tesseract executable
• also linux (g+++) cmake works

E.g. https://www.linuxtopia.org/online_books/an_introduction_to_gcc/gccintro_18.html (but also SO ) claims that:

$ gcc -Wall calc.c -lm -o calc   (correct order)
$ gcc -Wall -lm calc.c -o calc    (incorrect order)

so your propopsed compilation order is not correct.

[namniav]

Sorry that I messed up some memories about linkers... But the problem still exists and I have found the cause:

tesseract/src/viewer/svpaint.cpp

Line 230 in 82a7a87

int main(int argc, char** argv) {

This file is included by

tesseract/CMakeLists.txt

Line 406 in 82a7a87

src/viewer/*.cpp

I fed this patch to vcpkg and rebuilt tesseract, then the problem is fixed.

-add_library                     (libtesseract ${LIBRARY_TYPE} ${tesseract_src} ${arch_files}
+set(libtesseract_src ${tesseract_src})
+list(REMOVE_ITEM libtesseract_src
+                 ${CMAKE_CURRENT_SOURCE_DIR}/src/viewer/svpaint.cpp)
+
+add_library                     (libtesseract ${LIBRARY_TYPE} ${libtesseract_src} ${arch_files}

$ nm installed/x64-windows/lib/tesseract41.lib | grep "main"
... # no main function reported

My Qt&CMake project works fine now.

Also I guess the similar problem exists for tesseract executable target, since there are two main functions, one from src/api/tesseractmain.cpp and the other from src/viewer/svpaint.cpp. I have no idea how they could work together.

[amitdo]

svpaint.cpp seems to be a utility for testing ScrollView. It should not be compiled as part of libtesseract.

[amitdo]

The Autotools and sw build definitions do not contain an instruction to compile this file.

CMakeLists.txt does contain an instruction to compile it.

(

tesseract/CMakeLists.txt

Line 565 in a96b2ab

src/viewer/*.cpp

[amitdo]

An alternative to the suggested changes is to move svpaint.cpp to a subdirectory: viewer/test. Another one is to explicitly list file names instead of using *.cpp.

@stweil, do you have a preference?

[stweil]

Pull request #3873 should fix this issue. It moves src/viewer/svpaint.cpp to src/svpaint.cpp.

I also considered moving it to unittest, but think that the current solution is simpler.