Add Ruby 3.4 to linux CI

.github/workflows/test.yml

@@ -15,7 +15,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        xcode: ['16.0']
+        xcode: ['16.1']
     steps:
       - uses: maxim-lobanov/setup-xcode@v1
         with:
@@ -69,6 +69,7 @@ jobs:
           - short: '3.1'
           - short: '3.2'
           - short: '3.3'
+          - short: '3.4'
     steps:
       - uses: actions/checkout@v4
         with:

Sources/RubyGateway/RbVM.swift

@@ -109,6 +109,28 @@ final class RbVM : @unchecked Sendable {
             try RbError.raise(error: .setup("Has already been done (via C API?) for this process."))
         }
 
+        // What is going on with init_stack
+        // --------------------------------
+        // Line added for Ruby 3.4 because of ruby/ruby:9505 that took it out of `ruby_setup()`.
+        //
+        // This stack frame we're in now isn't very interesting except for defining the native thread
+        // that will become the Ruby "main thread".  Here is why it's OK to call this macro:
+        //
+        // `RUBY_INIT_STACK` declares a ‘stack’ variable and calls `vm.c:ruby_init_stack()` which
+        // stores that variable address in `native_main_thread_stack_top`, the only place that is set.
+        //
+        // `eval.c:ruby_setup()` -> `vm.c:Init_BareVM()` is the only place that refers to the static
+        // and passes to `thread.c:ruby_thread_init_stack()` for the current thread, which for platforms
+        // we care about[1] leads to `thread_pthread.c:native_thread_init_stack()`.  We only care about the
+        // “main thread” use-case at this point and go to`thread_pthread.c:native_thread_init_main_thread_stack()`.
+        // We only care about `MAINSTACKADDR_AVAILABLE` and so do not use the address to figure the
+        // stack layout for GC.  Then there is a sanity check which is the only place the address is
+        // used - as long as it is within the stack as reported by pthreads then we are good.
+        //
+        // [1] Brief eyeball the win32 version looks OK too.
+        //
+        rbg_RUBY_INIT_STACK()
+
         let setup_rc = ruby_setup()
         guard setup_rc == 0 else {
             try RbError.raise(error: .setup("ruby_setup() failed: \(setup_rc)"))

Sources/RubyGatewayHelpers/include/rbg_helpers.h

@@ -21,6 +21,9 @@
 typedef unsigned long VALUE;
 typedef VALUE ID;
 
+/// Call `RUBY_INIT_STACK`
+void rbg_RUBY_INIT_STACK(void);
+
 /// Safely call `rb_load` and report exception status.
 void rbg_load_protect(VALUE fname, int wrap, int * _Nonnull status);
 

Sources/RubyGatewayHelpers/rbg_macros.m

@@ -54,3 +54,8 @@ long rbg_RSTRING_LEN(VALUE v)
 // These become inlines in Ruby 3 that get imported
 int rbg_RB_TEST(VALUE v) { return RB_TEST(v); }
 int rbg_RB_NIL_P(VALUE v) { return RB_NIL_P(v); }
+
+// See comment on call in RbVM.swift
+void rbg_RUBY_INIT_STACK(void) {
+    RUBY_INIT_STACK;
+}

Tests/RubyGatewayTests/TestVM.swift

@@ -111,7 +111,8 @@ class TestVM: XCTestCase {
                 try Ruby.load(filename: Helpers.fixturePath("unloadable.rb"))
                 XCTFail("Managed to load unloadable file")
             } catch RbError.rubyException(let exn) {
-                XCTAssertTrue(exn.description.contains("SyntaxError:"))
+                let desc = exn.description
+                XCTAssertTrue(desc.hasPrefix("SyntaxError:") || desc.hasPrefix("NameError:"))
             }
         }
     }