Use kwargs with resolv-replace gem version >= 0.2.0

[jasonpenny]

When upgrading from v2 to latest, I hit a lot of errors about ruby/3.3.0/gems/timeout-0.4.3/lib/timeout.rb:95:in 'new': can't alloc thread (ThreadError) in a project that is using the resolv-replace bundled gem.

resolv-replace v0.2.0 updated its TCPSocket#initialize patch to forward **kwargs, which means connect_timeout: should now work correctly when that version is loaded. Previously, supports_connect_timeout? treated any resolv-replace-patched TCPSocket as incompatible.

What changed

supports_connect_timeout? now checks the parameter signature of TCPSocket#initialize more precisely instead of requiring an exact match against the native [[:rest]] signature:

• Unpatched Ruby 3.0+ – parameters are [[:rest]], returns true (unchanged)
• resolv-replace ≥ 0.2.0 – parameters include **kwargs (a :keyrest entry), returns true (new)
• resolv-replace < 0.2.0 / socksify / other patchers without kwargs – no :keyrest entry, returns false (unchanged)

This PR was generated with the assistance of claude code

[mattwd7]

🙏 🙏 🙏

[petergoldstein]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

🤖 Generated with Claude Code

[petergoldstein]

The JRuby CI failure on this PR is caused by a pre-existing issue that your change exposes more broadly: JRuby 10.1.0.0 (Ruby 4.0 compat) does not support TCPSocket.instance_method because JRuby's TCPSocket is Java-backed and doesn't expose Ruby method introspection.

Error:

NoMethodError: undefined method 'instance_method' for class TCPSocket
    lib/dalli/socket.rb:116:in 'supports_connect_timeout?'

The fix (being applied to main separately) is to guard with RUBY_ENGINE == 'ruby' — consistent with the test that already has skip 'MRI-specific test' if RUBY_ENGINE != 'ruby':

@supports_connect_timeout = RUBY_ENGINE == 'ruby' && RUBY_VERSION >= '3.0' &&
                            (params == TCPSOCKET_NATIVE_PARAMETERS || params.any? { |type, _| type == :keyrest })

This PR will need to incorporate that guard before it can merge. On JRuby, supports_connect_timeout? will return false and Dalli will fall back to Timeout.timeout, which is the correct behavior since connect_timeout: is MRI-specific functionality.

🤖 Generated with Claude Code

[petergoldstein]

@jasonpenny can you please rebase against main and make any other changes that may be required per my last comment? Thanks.

[jasonpenny]

@jasonpenny can you please rebase against main and make any other changes that may be required per my last comment? Thanks.

did you push a commit? I don't see it in GitHub (so this is still branched off of main)
I can apply the RUBY_ENGINE == 'ruby' update, but it sounds like you're going to do that separately?

[petergoldstein]

@jasonpenny Sorry, Claude didn't do the push - fixed now. Can you please rebase now and resolve? Thanks

[jasonpenny]

rebased, and I believe I resolved the RUBY_ENGINE issue

[petergoldstein]

Thanks!

[mattwd7]

@petergoldstein can we anticipate a 5.0.3 release anytime soon incorporating these changes?

[petergoldstein]

@mattwd7 Released