tensorflow-2.4.0 requires < GCC 10.3 Reproducible: Always Actual Results: >>> Configuring source in /var/tmp/portage/sci-libs/tensorflow-2.4.0/work/tensorflow-2.4.0 ... * python3_8: running run_in_build_dir do_configure * Only gcc version(s) 4.7 4.8 4.9 5.3 5.4 6.3 6.4 7.2 7.3 8.2 8.3 8.4 9.2 9.3 10.2 are supported, * of which none is installed * ERROR: sci-libs/tensorflow-2.4.0::gentoo failed (configure phase): * Only gcc version(s) 4.7 4.8 4.9 5.3 5.4 6.3 6.4 7.2 7.3 8.2 8.3 8.4 9.2 9.3 10.2 are supported Adding a BDEPEND in tensorflow-2.4.0.ebuild: <sys-devel/gcc-10.3.0 Fixes the problem.
With USE=cuda you need to manually emerge and select a viable gcc version yourself. DEPEND shouldn't really be used to try and control this. For example you may want gcc11 for your system, and gcc10.2 for cuda, but with <gcc-10.3 you can't have gcc11 anymore.
(In reply to Ionen Wolkens from comment #1) > With USE=cuda you need to manually emerge and select a viable gcc version > yourself. > > DEPEND shouldn't really be used to try and control this. For example you may > want gcc11 for your system, and gcc10.2 for cuda, but with <gcc-10.3 you > can't have gcc11 anymore. Ahh, I understand now. Yes, that sounds appropriate. This bug can now be closedas invalid.