The root 5 version that was in F24 before this update was completely broken. It had failed the F24 mass rebuild and was FTBFS. Root 5 is incompatible with gcc 6. The old cint c++ interpreter does not understand all the new features used in the system's g++ headers. This was not an issue about updating root to version 6 in F24 or not, it was a question about having root in F24 or not.
Yes, something went wrong with the push in EPEL 5 and EPEL 6, and the update with the older version was pushed later. It is a known problem that bodhi doesn't order different versions of the same package in the same push in increasing version order - so sometimes the wrong one gets pushed last.
When building packages in koji there is no network available, so the gen_include.sh script - which downloads source rpms using yumdownloader - can not be run. The globus_gfs_internal_hack.h header is therefore not re-generated during the build, but remains the version included in the sources.
The root.log lists the correct versions were installed for the build:
The build succeeded without code changes.
To check that it still builds today I just made scratch builds for rawhide and EPEL 6 which both succeeded without ny code changes.
When you say "in order to build the dpm-dsi with the new version of gridftp we had to make some code changes" does it mean that you had to make changes to make it compile, or it compiles without changes but changes are needed to make it work correctly?