Re: bad package build of gcc10

To: Jonathan Perkin <jperkin%mnx.io@localhost>
Subject: Re: bad package build of gcc10
From: Thomas Klausner <wiz%gatalith.at@localhost>
Date: Wed, 30 Oct 2024 12:46:49 +0100

On Wed, Oct 30, 2024 at 09:26:07AM +0000, Jonathan Perkin wrote:
> * On 2024-10-18 at 23:40 BST, Greg Troxel wrote:
> 
> > Rooting around destdir, I found a file that references quadmath:
> > 
> >  $ objdump -x work/.destdir/usr/pkg/gcc10/lib/libgfortran.so.5.0|egrep NEED\|RPATH
> >  NEEDED               libquadmath.so.0
> >  NEEDED               libm.so.0
> >  NEEDED               libgcc_s.so.1
> >  NEEDED               libc.so.12
> >  RPATH                /usr/pkg/gcc10//lib/.:/usr/pkg/gcc10/lib/.:/usr/pkg/gcc10/lib:/usr/pkg/lib
> >  VERNEED              0x0000000000017a98
> >  VERNEEDNUM           0x0000000000000002
> > 
> > and that seems to be the issue: //lib/. instead of /lib, leading to a
> > REQUIRES that should have been a match to the PROVIDES and hence
> > omittted.  Basically, the checking code is not using the same resolution
> > rules that the dynamic linker does.
> 
> I think what doesn't help is that people have been cargo-culting around GCC
> patches without understanding what they do.  It's one of the reasons I no
> longer use these packages.
> 
> At some point between gcc9 and gcc10, a very important part of
> patch-gcc_Makefile.in got corrupted, and it's been copied in its broken form
> ever since:
> 
> gcc6/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG)$(prefix)/@GCC_TARGET_MACHINE@/lib/%M $(LINKER_RPATH_FLAG)$(libdir)/%M \"" \
> gcc7/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG)$(prefix)/@MACHINE_GNU_PLATFORM@/lib/%M $(LINKER_RPATH_FLAG)$(libdir)/%M \"" \
> gcc8/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG)$(prefix)/@MACHINE_GNU_PLATFORM@/lib/%M $(LINKER_RPATH_FLAG)$(libdir)/%M \"" \
> gcc9/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG)$(prefix)/@MACHINE_GNU_PLATFORM@/lib/%M $(LINKER_RPATH_FLAG)$(libdir)/%M \"" \
> gcc10-aux/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG) $(prefix)//lib/%M $(LINKER_RPATH_FLAG) $(libdir)/%M \"" \
> gcc10/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG) $(prefix)//lib/%M $(LINKER_RPATH_FLAG) $(libdir)/%M \"" \
> gcc12/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG) $(prefix)//lib/%M $(LINKER_RPATH_FLAG) $(libdir)/%M \"" \
> gcc13-gnat/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG) $(prefix)//lib/%M $(LINKER_RPATH_FLAG) $(libdir)/%M \"" \
> gcc13/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG) $(prefix)//lib/%M $(LINKER_RPATH_FLAG) $(libdir)/%M \"" \
> gcc14/patches/patch-gcc_Makefile.in:+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG) $(prefix)//lib/%M $(LINKER_RPATH_FLAG) $(libdir)/%M \"" \
> 
> Note that the @MACHINE_GNU_PLATFORM@ substitution is no longer there, which
> probably explains your "//" problem, and likely is the cause of other
> issues.

Thanks.  The attached patch improves the rpath issue AFAICT, but the
REQUIRES problem is still there.

# readelf -a gcc10/x86_64--netbsd/lib/libgomp.so
...
 0x000000000000000f (RPATH)              Library rpath: [/usr/pkg/gcc10/x86_64--netbsd/lib/.:/usr/pkg/gcc10/lib/.:/usr/pkg/lib]
...

 Thomas

Index: gcc10/Makefile
===================================================================
RCS file: /cvsroot/pkgsrc/lang/gcc10/Makefile,v
retrieving revision 1.25
diff -u -r1.25 Makefile
--- gcc10/Makefile	1 Apr 2024 18:08:48 -0000	1.25
+++ gcc10/Makefile	30 Oct 2024 11:45:27 -0000
@@ -4,7 +4,7 @@
 ## When bumping the PKGREVISION of this package the PKGREVISION of
 ## lang/gcc10-libs needs to be bumped to be at least 1 more than the
 ## PKGREVISION of this package!
-PKGREVISION=	2
+PKGREVISION=	3
 
 COMMENT=	The GNU Compiler Collection (GCC) - 10.0 Release Series
 
@@ -50,6 +50,12 @@
 CONFIGURE_ARGS+=	--enable-libssp
 .endif
 
+SUBST_CLASSES+=		vars
+SUBST_MESSAGE.vars=	Setting target machine name path in ${SUBST_FILES.vars}
+SUBST_STAGE.vars=	pre-configure
+SUBST_FILES.vars=	gcc/Makefile.in
+SUBST_VARS.vars=	MACHINE_GNU_PLATFORM
+
 CONFIGURE_ENV.QNX+=	ACCEPT_INFERIOR_RM_PROGRAM=yes
 # Make invokes further configure scripts
 MAKE_ENV.QNX+=		ACCEPT_INFERIOR_RM_PROGRAM=yes
Index: gcc10/distinfo
===================================================================
RCS file: /cvsroot/pkgsrc/lang/gcc10/distinfo,v
retrieving revision 1.18
diff -u -r1.18 distinfo
--- gcc10/distinfo	1 Apr 2024 18:08:48 -0000	1.18
+++ gcc10/distinfo	30 Oct 2024 11:45:27 -0000
@@ -8,7 +8,7 @@
 Size (isl-0.16.1.tar.bz2) = 1626446 bytes
 SHA1 (patch-contrib_download__prerequisites) = 9cdcde21ab174052911447d20762ddfd40aa8791
 SHA1 (patch-fixincludes_inclhack.def) = 7b00974d3f52a8364190a607f52a5d7e8a6db41a
-SHA1 (patch-gcc_Makefile.in) = a47571d8340702d1a86d0dfb7df654280c6a6ee6
+SHA1 (patch-gcc_Makefile.in) = 9aa73d2234b1c9c63ea0f6d431ace82f666b82c6
 SHA1 (patch-gcc_config.gcc) = 4767d5b16794f4485879ed10e172cb2f86bc0777
 SHA1 (patch-gcc_config.host) = 8920ce841f6088b365c9b590e3ea9535bd880b40
 SHA1 (patch-gcc_config_aarch64_aarch64-builtins.c) = 87185f5c2e2dbe4195cb19fe6213d2d616d5519b
Index: gcc10/patches/patch-gcc_Makefile.in
===================================================================
RCS file: /cvsroot/pkgsrc/lang/gcc10/patches/patch-gcc_Makefile.in,v
retrieving revision 1.2
diff -u -r1.2 patch-gcc_Makefile.in
--- gcc10/patches/patch-gcc_Makefile.in	1 Apr 2024 14:33:57 -0000	1.2
+++ gcc10/patches/patch-gcc_Makefile.in	30 Oct 2024 11:45:27 -0000
@@ -11,7 +11,7 @@
  # Language-independent files.
  
  DRIVER_DEFINES = \
-+  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG) $(prefix)//lib/%M $(LINKER_RPATH_FLAG) $(libdir)/%M \"" \
++  -DLINK_LIBGCC_SPEC="\"%D $(LINKER_RPATH_FLAG)$(prefix)/@MACHINE_GNU_PLATFORM@/lib/%M $(LINKER_RPATH_FLAG)$(libdir)/%M \"" \
    -DSTANDARD_STARTFILE_PREFIX=\"$(unlibsubdir)/\" \
    -DSTANDARD_EXEC_PREFIX=\"$(libdir)/gcc/\" \
    -DSTANDARD_LIBEXEC_PREFIX=\"$(libexecdir)/gcc/\" \
Index: gcc10-libs/Makefile
===================================================================
RCS file: /cvsroot/pkgsrc/lang/gcc10-libs/Makefile,v
retrieving revision 1.9
diff -u -r1.9 Makefile
--- gcc10-libs/Makefile	1 Apr 2024 18:08:48 -0000	1.9
+++ gcc10-libs/Makefile	30 Oct 2024 11:45:27 -0000
@@ -9,7 +9,7 @@
 ## The PKGREVISION of this package needs to be at least 1 more than the
 ## PKGREVISION of the lang/gcc10 package so that with the dependence pattern
 ## '{gcc10,gcc10-libs}>=10.2.*' pkg_add will choose gcc10-libs over gcc10.
-PKGREVISION=	3
+PKGREVISION=	4
 
 CATEGORIES=	lang
 MASTER_SITES=	# empty

Follow-Ups:
- Re: bad package build of gcc10
  - From: Jonathan Perkin

References:
- false alert on new pkg_add shlibs check?
  - From: Greg Troxel
- Re: false alert on new pkg_add shlibs check?
  - From: Thomas Klausner
- Re: false alert on new pkg_add shlibs check?
  - From: Greg Troxel
- Re: false alert on new pkg_add shlibs check?
  - From: Thomas Klausner
- bad package build of gcc10
  - From: Greg Troxel
- Re: bad package build of gcc10
  - From: Jonathan Perkin

Prev by Date: Re: bad package build of gcc10
Next by Date: Re: bad package build of gcc10
Previous by Thread: Re: bad package build of gcc10
Next by Thread: Re: bad package build of gcc10
Indexes:

Home | Main Index | Thread Index | Old Index