No output with gcc 4.8.2

[PhilipDeegan]

OS: Centos 6.7 x86_64
Compiler: gcc version 4.8.2 20140120 (Red Hat 4.8.2-15)
Found: http://people.centos.org/tru/devtools-2/devtools-2.repo

Our compiler uses the system libc, so it could be that glibc is too old

ldd --version
ldd (GNU libc) 2.12

[dmah42]

i'd expect some sort of error at least... given i don't have a centos system to test on, can you run in a debugger and see what happens when you hit a print statement.

[PhilipDeegan]

The headers print showing time and iterations, but there's no results given for any tests, and there's no errors, at least printed errors.

[EricWF]

Could you please provide the benchmark your trying to run?

[PhilipDeegan]

`static void BM_StringCreation(benchmark::State& state) {
while (state.KeepRunning())
std::string empty_string;
}
// Register the function as a benchmark
BENCHMARK(BM_StringCreation);

// Define another benchmark
static void BM_StringCopy(benchmark::State& state) {
std::string x = "hello";
while (state.KeepRunning())
std::string copy(x);
}
BENCHMARK(BM_StringCopy);

BENCHMARK_MAIN();`

[EricWF]

I can't reproduce with GCC 4.8.5 on Ubuntu 16.04.

[PhilipDeegan]

It doesn't happen with GCC 4.9.2 / 5.2 / 6.2 or 7, only 4.8.2 with the RPMs noted.

No benchmarks run and it exits immediately after printing the headers.

[EricWF]

And your 100% sure it exits with a 0 exit code and doesn't produce any core dumps?

[EricWF]

I also can't reproduce with GCC 4.8.0 nor 4.8.2 on Ubuntu 16.04.

[EricWF]

What's the output of cat /opt/rh/devtoolset-2/enable?

[PhilipDeegan]

cat /opt/rh/devtoolset-2/enable

0 exit code, no coredumps.

[EricWF]

OK, I think I've figured this out. Did you compile libbenchmark.a using GCC 4.8.2, or more likely something like 5.2? Versions of the library built with the cxx11 ABI cannot be used with GCC < 5, and libraries built with the old ABI are not compatible with GCC > 5. I suspect if you build the library and test with the same compiler they will work.

However it's very weird that it's not causing a segfault.

[ctiller]

I'm seeing this too with grpc/grpc#8500.

[ctiller]

--benchmark_list_tests=true also outputs an empty list.

[PhilipDeegan]

Each time a different compiler is used libbenchmark is recompiled with all the same values as the main binary.

[dmah42]

i reproduced this with grpc. the regex is failing to match even when set to the default '.'

my guess is that grpc is building with make so we're not setting the regex machine correctly (we do some hoops in cmake to select it).

still debugging.