Issues with building LogDevice on Fedora Docker containers

[skindstrom]

I've found two issues with the Fedora docker containers, where both Fedora containers suffer from the same issue.

1. A crash occurs because gtest_discover_tests is not found. This is because this function was added to the GoogleTest cmake module in 3.10, and the Fedora docker containers use cmake 3.9.1.
   I suggest simply not using gtest_discover_tests when on a lesser version, as the tests seem to be added without this extra functionality anyways when GoogleTest can not be successfully loaded. I'll happily submit a PR for this.

2. The issue with _boost_py_component moved to Find boost-python reliably cross linux distributions #35

[calebmarchent]

Thanks for raising the issue, Simon. I can confirm this is a valid issue, needing to be fixed. I would propose that perhaps it is easier to move to using a more modern version of Fedora; coming pre-packaged with a newer cmake.

I would be tempted to split-off the _boost_py_component to a separate issue?

If you have solutions to the issue(s) you have raised, I would welcome your contribution.

[skindstrom]

I'm all for upgrading the Fedora version. Building on Fedora 28 results in an error, where folly redefines a struct from jemalloc. I could try to look into this at some later time.

/logdevice/external/folly/folly/experimental/JemallocHugePageAllocator.cpp:52:8: error: redefinition of 'struct extent_hooks_s'
 struct extent_hooks_s {
        ^~~~~~~~~~~~~~
In file included from /logdevice/external/folly/folly/memory/Malloc.h:32,
                 from /logdevice/external/folly/folly/experimental/JemallocHugePageAllocator.h:22,
                 from /logdevice/external/folly/folly/experimental/JemallocHugePageAllocator.cpp:17:
/usr/include/jemalloc/jemalloc.h:339:8: note: previous definition of 'struct extent_hooks_s'
 struct extent_hooks_s {
        ^~~~~~~~~~~~~~
make[5]: *** [CMakeFiles/folly_base.dir/build.make:882: CMakeFiles/folly_base.dir/folly/experimental/JemallocHugePageAllocator.cpp.o] Error 1
make[5]: *** Waiting for unfinished jobs....

If you feel as if the workaround in PR #32 is not wanted, I would suggest we set the required cmake version to 3.10 to fail early and verbosely, here:
https://github.com/facebookincubator/LogDevice/blob/2c0d9d5d9a38dfc5e60886acabc0a036eb43031e/logdevice/CMakeLists.txt#L7

[calebmarchent]

Update: Fedora 25 is out of support now, so no longer listed in popular package indices etc.
I have started looking into Fedora28 support today; the firstly by creating a LXC boxfile to work within: https://app.vagrantup.com/calebmarchent/boxes/fedora28-64-lxc; Will provide further updates when I've had a chance to look at the specific issues.

[calebmarchent]

I can confirm that I also see the jemalloc compile issue with the folly build on Fedora 28;
This would be the same issue as: facebook/folly#943

[calebmarchent]

I have tested the build with Fedora 27; it is successful (save the python boost detection issue #35; so while the folly team are already onto the jemalloc issue; I suggest we go for Fedora27 support.

[mickvav]

Fedora 29 builds fine (with same hackaround, I believe) - see #55

[calebmarchent]

Both the original reported issues have been addressed, many thanks @mickvav and @skindstrom ... the remaining work to support Fedora is: mstch for Fedora; and introduction of some CI job to keep it working.