rippled stops linking libgcc statically after std::call_once aborts
The rippled repository merged pull request 8721 on 19 November. It stops linking libgcc statically on Linux while keeping -static-libstdc++, removing -static-libgcc from the CMake link options, the Rust crates' build flags and the Conan profile. The CI check that verified static linking now only requires libstdc++ to be static.
The PR, by kuznetsss, describes the failure in detail. Throwing an exception from a std::call_once initializer is valid C++ and must reach the caller, but with both static flags the binary contained its own copy of the libgcc unwinder while glibc still loaded the shared libgcc_s.so.1 for its own frames. When an exception crossed a glibc frame with a cleanup handler, such as pthread_once, glibc called the shared copy on a context built by the static one, and the process aborted with no message. In CI this showed up as the test binary dying with SIGABRT, exit code 134, on every gcc job.
A second bug sits behind it: before glibc 2.34, pthread_once skipped its cleanup when a C++ exception passed through, so every later caller hung. A companion change, #8722, moves the Nix toolchain to glibc 2.34, and the PR notes it must land before or together with that change, because the glibc fix is what exposes the unwinder conflict. The author also points out that the "static" binaries already depended on libgcc_s.so.1 at run time without saying so.

Why it matters
Which symptom a validator operator would have seen, a hang or a silent abort, depended on the host's glibc, not on rippled's code. Making the dependency on the shared libgcc explicit trades a slightly less self-contained binary for one that behaves the same on every system.