bazel build somecpp_library fails if it depends on another library

[pmbethe09]

make the following:
cc_library(name="A", ...)
cc_library(name="B", srcs=["b.cpp"], deps=[":A"], ...)
cc_binary(name="main", ..., deps=[":B"])

bazel build :main works just fine (it correctly puts libA.so and libB.so on the link line)
bazel build :A - same
but bazel build :B will fail with:

From Linking lib/libB.so:
Undefined symbols for architecture x86_64:
"...."
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
ERROR: lib/BUILD:3:1: Linking of rule '//lib:lib' failed: gcc failed: error executing command /usr/bin/gcc -shared -o bazel-out/local_darwin-fastbuild/bin/lib/libB.so bazel-out/local_darwin-fastbuild/bin/lib/_objs/lib/lib/b.pic.o -lstdc++ -B/usr/bin -Wl,-S: com.google.devtools.build.lib.shell.BadExitStatusException: Process exited with status 1.

It appears that liba.so is missing from the command

[lberki]

This is very surprising. Couldn't reproduce with this BUILD file on Linux, though:

cc_library(
name = "a",
srcs = ["a.cpp"],
)

cc_library(
name = "b",
srcs = ["b.cpp"],
deps = [":a"],
)

cc_binary(
name = "main",
srcs = ["main.cpp"],
deps = [":b"],
)

Assigning to Damien since he knows a tad more about Mac than I do.

[damienmg]

Cannot reproduce:

    # Regression test for https://github.com/google/bazel/issues/67
    # C++ library depedending on C++ library fails to compile on Darwin
    function test_cpp_libdeps() {
        mkdir -p pkg
        cat <<'EOF' >pkg/BUILD
    cc_library(
      name = "a",
      srcs = ["a.cc"],
    )

    cc_library(
      name = "b",
      srcs = ["b.cc"],
      deps = [":a"],
    )

    cc_binary(
      name = "main",
      srcs = ["main.cc"],
      deps = [":b"],
    )
    EOF

        cat <<'EOF' >pkg/a.cc
    #include <string>

    std::string get_hello(std::string world) {
      return "Hello, " + world + "!";
    }
    EOF

        cat <<'EOF' >pkg/b.cc
    #include <string>
    #include <iostream>

    std::string get_hello(std::string);

    void print_hello(std::string world) {
      std::cout << get_hello(world) << std::endl;
    }
    EOF

        cat <<'EOF' >pkg/main.cc
    #include <string>
    void print_hello(std::string);

    int main() {
       print_hello(std::string("World"));
    }
    EOF

        bazel build //pkg:a >& $TEST_log \
            || fail "Failed to build //pkg:a"
        bazel build //pkg:b >& $TEST_log \
            || fail "Failed to build //pkg:b"
        bazel run //pkg:main >& $TEST_log \
            || fail "Failed to run //pkg:main"
        expect_log "Hello, World!";
    }

Succeed on my mac I will commit it to share it but the output is clean. I might have forgotten a flag though

EDIT NOTE: I added the two build to the test case to have a failing test case for future reference.

[damienmg]

\o/ I have a failing test

[pmbethe09]

Add an a.h which b.cc includes and calls a function from.
On Mar 27, 2015 6:45 AM, "lberki" [email protected] wrote:

This is very surprising. Couldn't reproduce with this BUILD file on Linux,
though:

cc_library(
name = "a",
srcs = ["a.cpp"],
)

cc_library(
name = "b",
srcs = ["b.cpp"],
deps = [":a"],
)

cc_binary(
name = "main",
srcs = ["main.cpp"],
deps = [":b"],
)

Assigning to Damien since he knows a tad more about Mac than I do.

—
Reply to this email directly or view it on GitHub
#67 (comment).

[damienmg]

The problem is that darwin clang expect to have all the dependency provided at compile time. It is not going to be fast to fix.

[damienmg]

I edited the test case in my previous message to have an actually failing one for future reference.

[kayasoze]

I've encountered this issue as well.

[aimeeble]

I've been hitting this too. I've worked around the issue by adding:

linker_flag: "-undefined"
linker_flag: "dynamic_lookup"

into my CROSSTOOL file. Not sure the larger implications of this style lookup though. But, it's enough for me to actually compile, so I'm happy for now.