Hi David, thank you for your continuing responses. I will try to get a backtrace with GTK debug symbols by patching Guix to add a debug output to GTK, in the hope that I patch the right GTK.
However, my believe is that the error is in G-Golf closures/callbacks/vfuncs (even though the hl-api tests do pass when run by Guix build or manually), specifically when I run valgrind on guile where I load and run drawing-widget.scm, it only shows errors from Boehm GC (at least mostly false-positive) and from creating threads, but no errors from GTK. In particular there are errors in scm_c_make_vector, probably false positive, but I will look at bytevectors this weekend now (occurrences of bv in g-golf code). I cannot tell you my host distro because there is none (there could be a foreign distro, but I have none). It is Guix System, that is Guix all the way down. To build G-Golf manually, I can and had patched the its g-golf/init.scm to use /gnu/store/xxxxxxx/lib/libgirepository-1.0.so file names like the guile-g-golf package at <https://git.savannah.gnu.org/cgit/guix.git/tree/gnu/packages/guile-xyz.scm> does, but it will not be useful. Setting the GTK_A11Y env var does not make a difference (except the warning is gone) and on my x86_64 GNOME setup there never is a warning (although GTK_A11Y is unset, but d-bus is there). Regards, Florian
