pkg/60763: lang/ruby34: does not link with the ruby-rjit option on NetBSD (rjit_c.o missing from DTRACE_DEPENDENT_OBJS)

["zakinko%snowrabbit.org@localhost via gnats" <gnats-admin%NetBSD.org@localhost>]

>Number:         60763
>Category:       pkg
>Synopsis:       lang/ruby34: does not link with the ruby-rjit option on NetBSD
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    pkg-manager
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Sep 22 08:45:00 +0000 2026
>Originator:     Showta Ishizaki
>Release:        NetBSD 11.0, pkgsrc trunk as of 2026-09-22
>Organization:
>Environment:
System: NetBSD 11.0 amd64
Architecture: x86_64
Machine: amd64
>Description:
	With PKG_OPTIONS.ruby34=ruby-rjit (instead of the default
	ruby-yjit), ruby34-3.4.10nb2 stops while linking miniruby:

	  ld: rjit_c.o: relocation R_X86_64_PLT32 against absolute symbol
	  `__dtraceenabled_ruby___cmethod__return' in section `.text' is disallowed

	rjit_c.c calls RUBY_DTRACE_CMETHOD_RETURN_HOOK, but rjit_c.o is
	not in DTRACE_DEPENDENT_OBJS in template/Makefile.in.  On NetBSD
	configure picks "dtrace needs post processing... rebuild", so
	"dtrace -G" rewrites only the objects in that list; rjit_c.o
	keeps the absolute placeholder symbols and ld refuses them.

	The default yjit build is not affected: rjit_c.c is compiled
	under #if USE_RJIT and is empty there.  Upstream fixed the same
	thing for yjit.o in 2022 (https://bugs.ruby-lang.org/issues/18480)
	and rjit_c.o was left out.  The ruby_3_4 branch still has it;
	master no longer has rjit_c.c, so this is a 3.4-only patch.

	The change is one line in the existing
	patches/patch-template_Makefile.in; the other two hunks are
	unchanged, only their offsets moved because the file was
	regenerated against 3.4.10.  I rebuilt with ruby-rjit
	on NetBSD 11.0/amd64: "processing probes in object files" runs,
	miniruby links, and

	  $ ruby34 -e 'puts RubyVM.const_defined?(:RJIT)'
	  true

>How-To-Repeat:
	  # cd /usr/pkgsrc/lang/ruby34
	  # make PKG_OPTIONS.ruby34=ruby-rjit build

	Fails at "linking miniruby" with the ld message above.

>Fix:
	patch -p0 from the top of pkgsrc.

	diff -ruN -x CVS -x work /usr/pkgsrc/lang/ruby34/Makefile lang/ruby34/Makefile
	--- /usr/pkgsrc/lang/ruby34/Makefile	2026-09-18 05:47:20.827063127 +0000
	+++ lang/ruby34/Makefile	2026-09-22 06:46:11.364176907 +0000
	@@ -2,7 +2,7 @@
	 
	 DISTNAME=	${RUBY_DISTNAME}
	 PKGNAME=	${RUBY_PKGPREFIX}-${RUBY_VERSION}
	-PKGREVISION=	2
	+PKGREVISION=	3
	 CATEGORIES=	lang ruby
	 MASTER_SITES=	${MASTER_SITE_RUBY}
	 
	diff -ruN -x CVS -x work /usr/pkgsrc/lang/ruby34/distinfo lang/ruby34/distinfo
	--- /usr/pkgsrc/lang/ruby34/distinfo	2026-09-18 05:47:20.831730759 +0000
	+++ lang/ruby34/distinfo	2026-09-22 06:46:11.686227894 +0000
	@@ -19,7 +19,7 @@
	 SHA1 (patch-lib_rubygems_installer.rb) = 4ef74b4f79837a929e81bcd0e7eba9061a442304
	 SHA1 (patch-lib_rubygems_platform.rb) = bde36a8fc1ba2fbf4d6fb8829bc116fb4d09b404
	 SHA1 (patch-lib_rubygems_specification.rb) = a3154185ef89fb33e699dd54b19a8e274f3275e0
	-SHA1 (patch-template_Makefile.in) = 3cf5d612ef725dc64d6a90d377b191a29ebfdf17
	+SHA1 (patch-template_Makefile.in) = 6138ccea7ac17dbb31a70d123700350154ffc503
	 SHA1 (patch-test_rubygems_test__gem.rb) = 32f7c7d7f8a024c045d78c2bce93944fc3113d04
	 SHA1 (patch-thread__pthread.c) = 7c1231933a2d6ce9d56891ab512371841697fbca
	 SHA1 (patch-tool_ifchange) = 1803bb6a1836e232dcabcf38f11c9881dbf726ea
	diff -ruN -x CVS -x work /usr/pkgsrc/lang/ruby34/patches/patch-template_Makefile.in lang/ruby34/patches/patch-template_Makefile.in
	--- /usr/pkgsrc/lang/ruby34/patches/patch-template_Makefile.in	2026-09-06 15:13:58.000000000 +0000
	+++ lang/ruby34/patches/patch-template_Makefile.in	2026-09-22 06:46:11.359395014 +0000
	@@ -8,9 +8,27 @@
	 
	 https://bugs.ruby-lang.org/issues/22059
	 
	+rjit_c.c has a DTrace hook (RUBY_DTRACE_CMETHOD_RETURN_HOOK) but its
	+object was not in DTRACE_DEPENDENT_OBJS, so on a DTRACE_REBUILD=yes
	+system such as NetBSD "dtrace -G" never processed it and linking
	+miniruby failed with --enable-rjit:
	+
	+  ld: rjit_c.o: relocation R_X86_64_PLT32 against absolute symbol
	+  `__dtraceenabled_ruby___cmethod__return' in section `.text' is disallowed
	+
	+Same fix as upstream made for yjit.o (https://bugs.ruby-lang.org/issues/18480).
	+
	 --- template/Makefile.in.orig	2026-05-07 20:25:14.644318083 +0000
	 +++ template/Makefile.in
	-@@ -523,7 +523,7 @@ probes.stamp: $(DTRACE_REBUILD_OBJS)
	+@@ -203,6 +203,7 @@
	+ 			string.$(OBJEXT) \
	+ 			symbol.$(OBJEXT) \
	+ 			vm.$(OBJEXT) \
	++			rjit_c.$(OBJEXT) \
	+ 			$(YJIT_OBJ)
	+ 
	+ THREAD_MODEL  = @THREAD_MODEL@
	+@@ -513,7 +514,7 @@
	  
	  probes.stamp: $(DTRACE_REBUILD_OBJS)
	  	$(Q) if test -f $@ -o -f probes.$(OBJEXT); then \
	@@ -19,7 +37,7 @@
	  	  $(ECHO0) "rebuilding objects which were modified by \"dtrace -G\""; \
	  	  $(MAKE) $(DTRACE_REBUILD_OBJS); \
	  	fi
	-@@ -537,7 +537,7 @@ probes.$(OBJEXT): $(srcdir)/probes.d $(DTRACE_REBUILD:
	+@@ -527,7 +528,7 @@
	  
	  # DTrace static library hacks described here:
	  # https://marc.info/?l=opensolaris-dtrace-discuss&m=114761203110734&w=4