This is my situation with everything I package (privately for my employer). I went in circles for a while simply believing I had to be doing something wrong until I considered the fact that most people doing packaging are not the authors. This all settled in completely when I began recalling the days of yore when one would download a tgz, extract, config, make, etc.. Still I think it's a shame that this isn't handled better. With very large projects it's quite a waste of time to archive just to meet the expected input format only to have the process reversed immediately.

That said, I do much as Igor has already mentioned. My build process starts with tito but lands in our Koji. I use the following Makefile without any changes for each of my projects to facilitate tito's tito.release.KojiGitReleaser:

$ cat Makefile
# Extract NVR from the spec while stripping any macros, specifically the
# disttag macro.
name := $(shell awk '/^Name:/{print $$2}' *.spec)
version := $(shell \
       awk '/^Version:/{print gensub(/%{.*?}/, "", "g", $$2)}' *.spec \
       )
release := $(shell \
       awk '/^Release:/{print gensub(/%{.*?}/, "", "g", $$2)}' *.spec \
       )
# The treeish we'll archive is effectively the Git tag that tito created.
treeish := ${name}-${version}-${release}

# Koji's buildSRPMFromSCM method expects a target named "sources" which
# ultimately must ensure that a tarball of the package's sources and its spec
# file are present. Our practice is to always keep a spec file in the Git
# repository, but we must build the tarball on the fly to resemble an upstream
# published work.
sources:
       git archive \
               --output=${name}-${version}.tar.gz \
               --prefix=${name}-${version}/ \
               ${treeish}