Repository navigation
glibc: lower the floor to 2.17 - #9
Merged
Merged
Conversation
The runtimes build with CMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY, so check_library_exists never links and always passes. libc++abi then referenced __cxa_thread_atexit_impl strongly, and any thread_local with a destructor failed to link on musl (already broken in 2026.10.0) and on glibc 2.17. Stage 30 and the add-on libc++ now set LIBCXXABI_HAS_CXA_THREAD_ATEXIT_IMPL=OFF there, using libc++abi's own fallback; hello.cpp covers it. Signed-off-by: Sam Gammon <sam@elide.dev>
Elide links against the gnu sysroot, so its 2.34 floor locked out RHEL/CentOS 7-8, Amazon Linux 2, Ubuntu 20.04 and Debian 11 (elide-dev/elide#1900). glibc 2.17 does not build with a current GCC or clang, so stage 20 now unpacks CentOS 7.9's final glibc, glibc-devel, glibc-headers and glibc-static RPMs (2.17-326.el7_9.3, x86_64 and AltArch aarch64, SHA256-pinned to the repodata digests) with a stdlib-only extractor, over stage 00's current UAPI headers. Libraries are flattened into /usr/lib like the old slibdir=/usr/lib build, with relative symlinks and /usr/lib linker-script paths. The glibc submodule and its CI fetch fallback are gone. Below 2.34, libpthread, libdl and librt are separate from libc: the gnu cfg links them --as-needed through link-only ($) options so C++ (libc++abi's pthread_key_create, libunwind's dladdr) links as before. The floor check, container checks (centos:7, amazonlinux:2, ubuntu:20.04), manifest (libcRevision = RPM release) and docs follow. TOOLCHAIN_VERSION 2026.10.1. Signed-off-by: Sam Gammon <sam@elide.dev>
The `$` cfg inputs are added to every link, including `clang -r`: libFuzzer's partial link of its private libc++ pulled static libpthread.a into fuzzer.o, which then failed to link (R_X86_64_32 against __stack_user). They were never needed: libc++, libc++abi and libunwind already name pthread and dl in .deplibs (C_SUPPORTS_COMMENT_LIB_PRAGMA), which lld honours. Signed-off-by: Sam Gammon <sam@elide.dev>
libc++abi and libunwind name pthread and dl in .deplibs, but under LTO the member that needs pthread_rwlock_* comes from bitcode and its deplibs never reach lld, so C programs linking the static unwinder with -flto failed (mimalloc's CMake probe in stage 35). The sysroot's libc.so script now lists AS_NEEDED(libpthread.so.0 libdl.so.2) and libpthread_nonshared.a ahead of libc.so.6 (the order an explicit -lpthread gives), so every link path (cfg, bare clang, GNU ld, rust-lld) resolves them and -r links are untouched. Signed-off-by: Sam Gammon <sam@elide.dev>
Sam Gammon (sgammon)
marked this pull request as ready for review
October 18, 2026 22:54
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Lowers the gnu sysroot's glibc from 2.34 to 2.17 on linux-amd64 and linux-arm64, so Elide (which links against this sysroot) runs on RHEL/CentOS 7-8, Amazon Linux 2, Ubuntu 20.04 and Debian 11 (elide-dev/elide#1900). Bumps
TOOLCHAIN_VERSIONto2026.10.1.Changes
glibc,glibc-devel,glibc-headers,glibc-static2.17-326.el7_9.3, x86_64 from vault7.9.2009/updates, aarch64 from AltArch). Each RPM is pinned inversions.envby its SHA256, cross-checked against the repodata digests.scripts/rpm-extract.pyunpacks them using only the Python stdlib, on top of stage 00's current UAPI headers; CentOS's 3.10 kernel-headers aren't used. Everything is flattened into/usr/libwith relative symlinks, and linker scripts point at/usr/lib, the same layout as before. Theglibcsubmodule and the CI sourceware/mirror fallback are removed..deplibsdon't reach lld under LTO. The sysroot'slibc.solinker script now listsAS_NEEDED(libpthread.so.0 libdl.so.2)andlibpthread_nonshared.aahead oflibc.so.6, which is the order an explicit-lpthreadgives. As a result-lcbehaves as it does on 2.34+, on every link path (cfg, bare clang, GNU ld, rust-lld), and-rlinks are left alone.librtusers must still add-lrt.STATIC_LIBRARYtry-compiles makecheck_library_existsalways pass. As a result libc++abi took a strong reference to__cxa_thread_atexit_impl, which glibc only has from 2.18 and musl doesn't have at all. Athread_localwith a destructor therefore failed to link on glibc 2.17, and it already fails on musl in 2026.10.0. The fix setsLIBCXXABI_HAS_CXA_THREAD_ATEXIT_IMPL=OFFfor both, andhello.cppnow covers this case.GLIBC_2.17. Container checks now run oncentos:7,amazonlinux:2andubuntu:20.04. The manifest'slibcRevisionis the RPM release.aarch64 note: upstream glibc 2.17 on aarch64 exports
_mcount@@GLIBC_2.18. Only-pgcode references it, and the floor check would still catch any such reference.No release, tag or upload is part of this PR.