#1605 fmt custom namespace

[jstrzebonski]

Fixes #1605

---

fmt has macros FMT_BEGIN_NAMESPACE and FMT_END_NAMESPACE, but unfortunately, internally it explicitly uses full namespace (in core.h):

namespace detail {
void to_string_view(...);
using fmt::v7::to_string_view;
...

To use it, we would have to hardcode our namespace anyway.

To solve this I extracted this "inner" namespace into an additional macro, so it could be simply defined in cmake.

---

I opened an issue in fmt repo regarding this problem
fmtlib/fmt#2642

---

Everything above is no longer valid. fmt team remove qualifying by inline namespace from core.h.

I copied their fix into our fmt/core.h and define FMT_BEGIN_NAMESPACE and FMT_END_NAMESPACE.

[github-actions]

PR tests (clang-3.9, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (clang-5.0, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (gcc-5, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (gcc-6, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (gcc-9, ubuntu, mpich, zoltan)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (gcc-7, ubuntu, mpich, trace runtime, LB)

Build for 8c07571

FAILED: tests/message_basic 
: && /usr/bin/g++ -O3 -DNDEBUG --coverage -rdynamic -Wl,-rpath -Wl,/usr/local/lib -Wl,--enable-new-dtags tests/CMakeFiles/message_basic.dir/Unity/unity_0_cxx.cxx.o -o tests/message_basic  -Wl,-rpath,/usr/local/lib  lib/brotli/libbrotlicommon-static.a  lib/brotli/libbrotlienc-static.a  lib/brotli/libbrotlidec-static.a  lib/libfort/lib/libfort.a  lib/libgtest.a  src/libvt-release.a  -ldl  /usr/lib/x86_64-linux-gnu/libz.so  lib/fmt/libfmt.a  /build/checkpoint/install/lib/libcheckpoint.a  -lpthread  lib/brotli/libbrotlienc-static.a  lib/brotli/libbrotlidec-static.a  lib/brotli/libbrotlicommon-static.a  -lm  lib/libfort/lib/libfort.a  /usr/local/lib/libmpicxx.so  /usr/local/lib/libmpi.so && :
collect2: error: ld returned 1 exit status
FAILED: tests/pipe_basic 
: && /usr/bin/g++ -O3 -DNDEBUG --coverage -rdynamic -Wl,-rpath -Wl,/usr/local/lib -Wl,--enable-new-dtags tests/CMakeFiles/pipe_basic.dir/Unity/unity_0_cxx.cxx.o -o tests/pipe_basic  -Wl,-rpath,/usr/local/lib  lib/brotli/libbrotlicommon-static.a  lib/brotli/libbrotlienc-static.a  lib/brotli/libbrotlidec-static.a  lib/libfort/lib/libfort.a  lib/libgtest.a  src/libvt-release.a  -ldl  /usr/lib/x86_64-linux-gnu/libz.so  lib/fmt/libfmt.a  /build/checkpoint/install/lib/libcheckpoint.a  -lpthread  lib/brotli/libbrotlienc-static.a  lib/brotli/libbrotlidec-static.a  lib/brotli/libbrotlicommon-static.a  -lm  lib/libfort/lib/libfort.a  /usr/local/lib/libmpicxx.so  /usr/local/lib/libmpi.so && :
collect2: error: ld returned 1 exit status
FAILED: tests/CMakeFiles/collection_extended.dir/Unity/unity_0_cxx.cxx.o 
/usr/bin/ccache /usr/bin/g++ -DFMT_BEGIN_NAMESPACE="namespace fmt { inline namespace vt {" -DFMT_END_NAMESPACE=}} -DJSON_USE_IMPLICIT_CONVERSIONS=1 -DVT_NO_COLOR_ENABLED -I/vt/tests/unit -I/vt/lib/CLI -I/vt/lib/json/include -I/vt/lib/brotli/c/include -I/vt/lib/libfort/lib -Irelease -I/vt/src -isystem /vt/tests/extern/googletest/googletest/include -isystem /vt/tests/extern/googletest/googletest -isystem /vt/lib/fmt/include -isystem /build/checkpoint/install/include -isystem /build/detector/install/include -O3 -DNDEBUG -fdiagnostics-color=always -Wall -pedantic -Wshadow -Wno-unknown-pragmas -Wsign-compare -ftemplate-backtrace-limit=100 -Werror -O0 -g --coverage -fPIC -std=c++14 -MD -MT tests/CMakeFiles/collection_extended.dir/Unity/unity_0_cxx.cxx.o -MF tests/CMakeFiles/collection_extended.dir/Unity/unity_0_cxx.cxx.o.d -o tests/CMakeFiles/collection_extended.dir/Unity/unity_0_cxx.cxx.o -c tests/CMakeFiles/collection_extended.dir/Unity/unity_0_cxx.cxx

Build log

[github-actions]

PR tests (gcc-10, ubuntu, openmpi, no LB)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (nvidia cuda 11.0, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (clang-9, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (nvidia cuda 10.1, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (clang-10, alpine, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (clang-10, ubuntu, mpich)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[github-actions]

PR tests (gcc-8, ubuntu, mpich, address sanitizer)

Build for 9cb5849

Compilation - successful

Testing - passed

Build log

[jstrzebonski]

There are some nvcc errors to be fixed.

[jstrzebonski]

It seems that nvcc is unable to handle definitions with spaces. I also observed that definition defined within vt's cmake was "leaked" into my test app, where I wanted to use different fmt.

Currently I just hardcoded vt namespace in fmt code. It's just two places in core.h that needed modification, and it fixes both problems.

[cz4rs]

Currently I just hardcoded vt namespace in fmt code. It's just two places in core.h that needed modification, and it fixes both problems.

Looks good to me, imho that's acceptable "manual" maintenance burden.

[jstrzebonski]

Currently I just hardcoded vt namespace in fmt code. It's just two places in core.h that needed modification, and it fixes both problems.

Looks good to me, imho that's acceptable "manual" maintenance burden.

Yes, in case of fmt update, that's literally two characters that need to be changed.