Where are the gcov symbols?

I'm trying to compile a simple app with gcov and getting the following link errors:

gcc AllTests.o CuTestTest.o CuTest.o -o TestTest
AllTests.o: In function `global constructors keyed to 0_RunAllTests':
/home/p7539c/cutest/AllTests.c:26: undefined reference to `__gcov_init'
AllTests.o:(.data+0x44): undefined reference to `__gcov_merge_add'
CuTestTest.o: In function `global constructors keyed to 0_TestCuStringNew':
/home/p7539c/cutest/CuTestTest.c:30: undefined reference to `__gcov_init'
CuTestTest.o:(.data+0x64): undefined reference to `__gcov_merge_add'
CuTest.o: In function `global constructors keyed to 0_CuStrAlloc':
/home/p7539c/cutest/CuTest.c:379: undefined reference to `__gcov_init'
CuTest.o:(.data+0x184): undefined reference to `__gcov_merge_add'
collect2: ld returned 1 exit status
make: *** [TestTest] Error 1

I can't seem to find the location of the missing symbols. gcov is present on the machine running gcc version 4.1.2

Any ideas? Thanks.

On Edit:

Everything seems to work fine when using gcov with an application that consists of one .c file. When I have multiple .c files (hence multiple .o files) I have the above problem.

The compile steps look like the following:

cc -fprofile-arcs -ftest-coverage -g   -c -o AllTests.o AllTests.c
cc -fprofile-arcs -ftest-coverage -g   -c -o CuTestTest.o CuTestTest.c
cc -fprofile-arcs -ftest-coverage -g   -c -o CuTest.o CuTest.c

I just spent an incredible amount of time debugging a very similar error. Here's what I learned:

• You have to pass -fprofile-arcs -ftest-coverage when compiling.

• You have to pass -fprofile-arcs when linking.

• You can still get weird linker errors when linking. They'll look like this:

  libname.a(objfile.o):(.ctors+0x0): undefined reference to 'global constructors keyed to long_name_of_file_and_function'

This means that gconv's having problem with one of your compiler-generated constructors (in my case, a copy-constructor). Check the function mentioned in the error message, see what kinds of objects it copy-constructs, and see if any of those classes doesn't have a copy constructor. Add one, and the error will go away.

Edit: Whether or not you optimize can also affect this. Try turning on / switching off optimizations if you're having problems with it.

The flag you're looking for is -lgcov when linking. That is, change:

gcc AllTests.o CuTestTest.o CuTest.o -o TestTest

to

gcc -lgcov AllTests.o CuTestTest.o CuTest.o -o TestTest

I found, as suggested here, that adding -lgcov to the build line when building a shared library that contains .o's built with -fprofile-arcs -ftest-coverage solved this for me. And of course linking the executable with -lgcov. Built the shared library like so:

g++    -shared -o libMyLib.so src_a.o src_b.o src_c.o -lgcov

And the executable like so:

g++ -o myExec myMain.o -lMyLib -lgcov

Adding the -lgov to the build of the shared library (not just the exe), solved this additional error for me:

hidden symbol `__gcov_merge_add' in /usr/lib/gcc/x86_64-redhat-linux/4.1.2/libgcov.a(_gcov_merge_add.o) is referenced by DSO
/usr/bin/ld: final link failed: Nonrepresentable section on output

Please note that -lgcov must be the last linked library.

You should be able to specify only --coverage on the command line when compiling and linking.

According to man gcc:

> The option is a synonym for -fprofile-arcs -ftest-coverage (when > compiling) and -lgcov (when linking).

I tried a simple test file with gcc -ftest-coverage -fprofile-arcs test.c and had no problems like you describe.

I suspect that gcc brings in the gcov library if the -ftest-coverage flag is there when it is linking. Try passing that flag on your gcc command line.

Perhaps obviously, this exact error message is produced when linking with a non-gcc linker. We seeing this error when linking with ifort (because our code includes both Fortran and C modules). Switching to linking with gcc did the trick.

Great Max Lybbert, Basically in case of autoconf usage add _LDADD = -lgcov ...

This will solve the issue.

So I added -shared to the CFLAGS, and now it seems to work with multiple files. Of course, it's coring in a strange place, so I don't know what that's about yet.