https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100486
--- Comment #63 from Arnaud Charlet <charlet at gcc dot gnu.org> --- (In reply to Eric Botcazou from comment #62) > > No need to, there are actually several reasons to not rely on exception > > propagation during bootstrap, so let's restore this invariant instead, I'm > > testing a change to that effect. > > Fair enough. In any case, the symptoms are again consistent with a lack of > registration of unwind tables, this time for libgcc, so this might be the > same underlying issue as the grep-induced mess for GCC 10. My hunch would > point to a constructor section merging issue in the linker, maybe related > to the GCC configure check for HAVE_LD_RO_RW_SECTION_MIXING, which should > fail for MinGW and may incorrectly succeed when things go wrong. Right, that's most likely the underlying issue and indeed ensuring we don't use exception propagation during the bootstrap is just a way to hide/avoid this underlying mismatch.
