On Fri, 2020-06-19 at 16:57 +0100, Daniel P. Berrangé wrote: This will not work for a few reasons: * you forgot "test" in the second part of the if condition, so the shell will tell you either skip: command not found or : command not found and not actually run rpmbuild; * you're using '-ta', but since the argument you're passing to rpmbuild in this case is not a tarball, you'll get error: Failed to read spec file from mingw-libvirt.spec I believe you need to use '-ba' instead; * as you mention in the commit message, the spec file is set up to build both mingw32 and mingw64 RPMs, but each of the cross-build containers only has dependencies for the corresponding architecture, so the build will fail anyway. This last one is the only one that's not trivially addressed. The standard RPM macros for MinGW already have built-in support for only building one of the two variants of the packages[1], but our mingw.spec files are not prepared for it: if I try to build, for example, libvirt-glib with $ rpmbuild --define "_sourcedir $PWD" \ --define "mingw_build_win32 0" \ -ba mingw-libvirt-glib.spec it eventually fails because, unsurprisingly, it can't find some mingw32 files. So we need to change all our mingw.spec files to support this properly before we can enable MinGW RPM build jobs. Since that's a bunch of extra work, and we never had MinGW RPM build jobs on CentOS CI, I suggest we skip that for now: let's focus on getting all projects on GitLab CI with more or less the same set of build jobs they had on CentOS CI, and once we've done that we can go back and add extra features such as MinGW RPMs. Does that sound reasonable? [1] https://fedoraproject.org/wiki/Packaging:MinGW#Build_for_multiple_targets -- Andrea Bolognani / Red Hat / Virtualization