On Tue, Oct 19, 2021 at 11:34 AM Martin Liška <mli...@suse.cz> wrote: > > On 10/19/21 11:12, Richard Biener wrote: > > On Fri, Oct 15, 2021 at 5:22 PM Martin Liška <mli...@suse.cz> wrote: > >> > >> All right, and there's second part that moves the code > >> from toplev.c to opts.c (finish_options) as I've done in the original > >> version. > >> > >> The patch also handles PR102766 where nvptx.c target sets: > >> debug_nonbind_markers_p = 0; > >> > >> So the easiest approach is marking the flag as set in global_options_set, > >> I haven't found a better approach :/ Reason is that nvptx_option_override > >> is called before finish_options. > > > > So currently nvptx_option_override is called before we do this code > > blob (it's called at the beginning of process_options). > > Yes, happens early in process_options. > > > > > Why's the solution not to move this setting to finish_options as well? > > I would like to, but the option detection depends on target hooks and some > debuginfo > functionality, leading to: > > g++ -no-pie -g -DIN_GCC -fPIC -DCROSS_DIRECTORY_STRUCTURE > -fno-exceptions -fno-rtti -fasynchronous-unwind-tables -W -Wall > -Wno-narrowing -Wwrite-strings -Wcast-qual -Wno-error=format-diag > -Wmissing-format-attribute -Woverloaded-virtual -pedantic -Wno-long-long > -Wno-variadic-macros -Wno-overlength-strings -fno-common -DHAVE_CONFIG_H > -static-libstdc++ -static-libgcc -o Tlto-wrapper \ > lto-wrapper.o collect-utils.o ggc-none.o libcommon-target.a libcommon.a > ../libcpp/libcpp.a ../libbacktrace/.libs/libbacktrace.a > ../libiberty/pic/libiberty.a ../libdecnumber/libdecnumber.a > /home/marxin/Programming/gcc/gcc/opts.c:1382: error: undefined reference to > 'dwarf2out_default_as_loc_support()' > /home/marxin/Programming/gcc/gcc/opts.c:1384: error: undefined reference to > 'dwarf2out_default_as_locview_support()' > /home/marxin/Programming/gcc/gcc/opts.c:1409: error: undefined reference to > 'targetm' > > Or can we do better?
Meh ... :/ Well, move the target override hook call down (try to shuffle things so diagnostics happen after but "inits" happen before). > > (and disabling it along var-tracking when we end with no -g in > > process_options) > > > > IMHO the target should have the last say, so the hook should be invoked > > after we are finished overriding stuff and finish_options should be > > _only_ doing diagnostics and disabling stuff we cannot handle. > > Well, that sounds good, but we are quite far from that :/ Yes, I know... > Martin > > > > > Richard. > > > >> Patch can bootstrap on x86_64-linux-gnu and survives regression tests. > >> > >> Ready to be installed? > >> Thanks, > >> Martin >
