Re: [Libguestfs] [PATCH 6/7] python: stop including config.h

On Friday, 10 January 2020 23:13:35 CET Richard W.M. Jones wrote: That is the thing: config.h is not created at all when building the Python binding (either as part of libguestfs, or when building the standalone tar.gz dist), but instead the one from libguestfs is used. Actually, it happens to _not_ work right now: https://bugzilla.redhat.com/show_bug.cgi?id=1627964 the libguestfs config.h is shipped in the tarball, and the content of that file depends on the way the binding was configured from the libguestfs build that created it. So in case of that bug, the tarball was created with Python 2. And while patches #3, #4, and #5 should fix the majority of the visible cases of this issue, there are still potential issues: config.h defines a series of macros (_GNU_SOURCE, __EXTENSIONS__, etc) that change the way things are built, and strictly depend on the platform. I'll reverse the question: if building the standalone tarball works fine with no config.h, why shouldn't the build-in-libguestfs do the same? The bindings do not use gnulib and use only few system headers and APIs, and nothing of the config.h content really matters (the closest thing is HAVE_PYTHON, which is obviously true...). In the end, it is just a binding that wraps the C API for a different language (Python in this case), so it definitely _ought to_ not depend on the way libguestfs itself was configured/built. Moreover: the same situation applies to the other bindings too (those with C parts, of course). -- Pino Toscano