https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128870
--- Comment #5 from Filip Sajdak <filip.sajdak at gmail dot com> --- More data for the file in the description, which ties it to the reproducer of comment 4. GCC 16.2.0 (Homebrew build, aarch64-apple-darwin25), g++-16 -std=c++23 <flags> -c red4.cpp, failures out of 30 runs each: -Os 19 -Os -fno-declone-ctor-dtor 0 -Oz -fno-declone-ctor-dtor 0 -O2 0 -O2 -fdeclone-ctor-dtor 16 So the trigger is -fdeclone-ctor-dtor (which -Os and -Oz enable) and not the size level as such, and -fno-declone-ctor-dtor works around it here. That is the same flag as in PR 125359 comment 8.
