update gc to version 7.6

[dimpase]

An update to gc is needed to support arm64/aarch64, see #23687, as well as for FreeBSD support, see #22679. It also appears to be good to be in sync with Linux boxes that ship make with guile extension (the latter depends on gc, and causes interesting errors, see #24575).

The source is here: ​https://github.com/ivmai/bdwgc/releases/download/v7.6.4/gc-7.6.4.tar.gz

On some platforms this new version also needs libatomic_ops, see #23701

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​7d09a89	libatomic_ops added as standard pkg
​7c8c3e3	Merge branch 'libatomic_ops' into gcupdate
​277ddd7	libatomic_ops is a dependency

[dimpase]

Erik, could you please check whether this works on Cygwin without your patch?

(one need #23701 first, and re-running ./configure to register the new package)

[embray]

Ah, yes. I remember this. I'll give it a go.

[embray]

I think this branch needs to be updated on the latest version of #23701 since it's using 7.4.6 and this branch is using 7.4.4.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​4d5bbd0	libatomic_ops 7.4.6 added as standard pkg
​e53f784	Merge branch 'u/dimpase/libatomic_ops' of trac.sagemath.org:sage into gcupdate

[dimpase]

You're right - done now.

[embray]

Thanks--now it's working. Or at least, it's building. Can't say anything about tests right now.

[dimpase]

on FreeBSD this needs the patch fixing ​https://github.com/ivmai/bdwgc/issues/132

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​8166e21	libatomic_ops 7.6.4 added as standard pkg
​64da17f	Merge branch 'libatomic_ops' into gcupdate
​82eabae	patch for FreeBSD
​931652a	Merge branch 'u/dimpase/gcupdate' of trac.sagemath.org:sage into gcupdate

[dimpase]

On Linux tests pass, but I can trigger a nasty crash interactively:

sage: from sage.libs.ecl import *
sage: from sage.interfaces.maxima_lib import <TAB>

Debian ​experimental uses gc 7.4.4, so this would be one to test for this regression, too.

[dimpase]

With GC included in ECL (tagged version 7.5.0 in ECL source, although I haven't checked how canonical this is) I do get the crash in comment 11.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​f9ca589	Merge branch 'u/dimpase/gcupdate' of trac.sagemath.org:sage into gc76
​e365c32	backported patch for FreeBSD support

[embray]

Turns out this is still not building properly on Cygwin. It fails to produce a shared lib at link time:

/bin/sh ./libtool  --tag=CC   --mode=link gcc   -fexceptions -Wall -Wextra -O2 -g  -fno-strict-aliasing  -version-info 1:3:0 -no-undefined -L/home/embray/src/sagemath/sage/local/lib -Wl,-rpath,/home/embray/src/sagemath/sage/local/lib  -o libgc.la -rpath /home/embray/src/sagemath/sage/local/lib allchblk.lo alloc.lo blacklst.lo checksums.lo dbg_mlc.lo dyn_load.lo finalize.lo gc_dlopen.lo gcj_mlc.lo headers.lo mach_dep.lo malloc.lo mallocx.lo mark.lo mark_rts.lo misc.lo new_hblk.lo obj_map.lo os_dep.lo pcr_interface.lo ptr_chck.lo real_malloc.lo reclaim.lo specific.lo stubborn.lo thread_local_alloc.lo typd_mlc.lo win32_threads.lo     fnlz_mlc.lo    -L/home/embray/src/sagemath/sage/local/lib -latomic_ops
mv -f cord/.deps/libcord_la-cordxtra.Tpo cord/.deps/libcord_la-cordxtra.Plo

*** Warning: linker path does not have real file for library -latomic_ops.
*** I have the capability to make that library automatically link in when
*** you link to this library.  But I can only do this if you have a
*** shared version of the library, which you do not appear to have
*** because I did check the linker path looking for a file starting
*** with libatomic_ops and none of the candidates passed a file format test
*** using a file magic. Last file checked: /home/embray/src/sagemath/sage/local/lib/libatomic_ops.a
*** The inter-library dependencies that have been dropped here will be
*** automatically added whenever a program is linked with this library
*** or is declared to -dlopen it.

*** Since this library must not contain undefined symbols,
*** because either the platform does not support them or
*** it was explicitly requested with -no-undefined,
*** libtool will only create a static version of it.
libtool: link: ar cru .libs/libgc.a  allchblk.o alloc.o blacklst.o checksums.o dbg_mlc.o dyn_load.o finalize.o gc_dlopen.o gcj_mlc.o headers.o mach_dep.o malloc.o mallocx.o mark.o mark_rts.o misc.o new_hblk.o obj_map.o os_dep.o pcr_interface.o ptr_chck.o real_malloc.o reclaim.o specific.o stubborn.o thread_local_alloc.o typd_mlc.o win32_threads.o fnlz_mlc.o
libtool: link: ranlib .libs/libgc.a
libtool: link: ( cd ".libs" && rm -f "libgc.la" && ln -s "../libgc.la" "libgc.la" )
/bin/sh ./libtool  --tag=CC   --mode=link gcc   -fexceptions -Wall -Wextra -O2 -g  -fno-strict-aliasing -version-info 1:3:0 -no-undefined -L/home/embray/src/sagemath/sage/local/lib -Wl,-rpath,/home/embray/src/sagemath/sage/local/lib  -o libcord.la -rpath /home/embray/src/sagemath/sage/local/lib cord/libcord_la-cordbscs.lo cord/libcord_la-cordprnt.lo cord/libcord_la-cordxtra.lo ./libgc.la

*** Warning: This system cannot link to static lib archive ./libgc.la.
*** I have the capability to make that library automatically link in when
*** you link to this library.  But I can only do this if you have a
*** shared version of the library, which you do not appear to have.

*** Warning: linker path does not have real file for library -latomic_ops.
*** I have the capability to make that library automatically link in when
*** you link to this library.  But I can only do this if you have a
*** shared version of the library, which you do not appear to have
*** because I did check the linker path looking for a file starting
*** with libatomic_ops and none of the candidates passed a file format test
*** using a file magic. Last file checked: /home/embray/src/sagemath/sage/local/lib/libatomic_ops.a
*** The inter-library dependencies that have been dropped here will be
*** automatically added whenever a program is linked with this library
*** or is declared to -dlopen it.

*** Since this library must not contain undefined symbols,
*** because either the platform does not support them or
*** it was explicitly requested with -no-undefined,
*** libtool will only create a static version of it.
libtool: link: ar cru .libs/libcord.a  cord/libcord_la-cordbscs.o cord/libcord_la-cordprnt.o cord/libcord_la-cordxtra.o
libtool: link: ranlib .libs/libcord.a
libtool: link: ( cd ".libs" && rm -f "libcord.la" && ln -s "../libcord.la" "libcord.la" )

It's probably a minor issue with link flag ordering. I will investigate and provide a solution.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​7b0f626	Merge branch 'u/dimpase/libatomic_ops' of trac.sagemath.org:sage into laup
​5c5d8f6	remove meaningless flags
​32b8eeb	Merge branch 'u/dimpase/gcupdate' of trac.sagemath.org:sage into laup

[dimpase]

updated the branch to take into account update on #23701

[embray]

Ah, I should have warned you, I'm about to propose another needed change for #23701.

[dimpase]

We should go for 7.6.2: ​https://github.com/ivmai/bdwgc/releases/tag/v7.6.2 In particular, this includes the patches needed by the FreeBSD port. I will update this accordingly.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​54e1da7	Merge branch 'develop' into libatomic_ops
​a024f8d	7.6.2 is the current version
​7db1b14	Merge 'libatomic_ops' -> gcupdate, update to 7.6.2

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​b97bf99

removing already included patches

[git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

​e790e65

removing already included patches

[dimpase]

OK, another advantage is that all our patches are now upstream!

[embray]

I'm trying to get this built on Cygwin--will report back with any findings.

[embray]

IIRC I had some trouble with this on Cygwin (specifically with libatomic_ops), but that was a while ago and I got distracted. Going to give it another try.

[embray]

Likewise with ticket:23701#comment:15 would you accept a small update to this for Cygwin?

[dimpase]

sure

[embray]

Rebased on top of my updates to #23701 and incorporated a minor change for Cygwin.

---

New commits:

​7135b2f	On Cygwin install libatomic_ops as a shared lib only; this is inline with what the official Cygwin package does
​18813a2	Convert libatomic_ops to use sage-dist-helpers
​e57cfc4	update to 7.6.0
​897c091	patch for FreeBSD
​2d923f2	backported patch for FreeBSD support
​822cc06	removing already included patches
​f6fd533	update gc to 7.6.2
​c2eb88c	For Cygwin only builds shared libraries

[vbraun]

Found local metadata for gc-7.6.2
Using cached file /mnt/disk/home/release/Sage/upstream/gc-7.6.2.tar.gz
gc-7.6.2
====================================================
Setting up build directory for gc-7.6.2
Finished extraction
Applying patches from ../patches...
Applying ../patches/0001-On-Cygwin-use-mprotect-instead-of-mmap-to-set-PROT_N.patch
patching file os_dep.c
Hunk #1 FAILED at 698.
Hunk #2 FAILED at 2404.
Hunk #3 FAILED at 2515.
3 out of 3 hunks FAILED -- saving rejects to file os_dep.c.rej
Error applying '../patches/0001-On-Cygwin-use-mprotect-instead-of-mmap-to-set-PROT_N.patch'
************************************************************************
Error applying patches
************************************************************************
Please email sage-devel (http://groups.google.com/group/sage-devel)
explaining the problem and including the log file
  /mnt/disk/home/release/Sage/logs/pkgs/gc-7.6.2.log
Describe your computer, operating system, etc.
************************************************************************

[dimpase]

facepalm... we should bump it up to 7.6.4 anyway.

[dimpase]

I believe that the failing patch is already in 7.6.4, see ​https://github.com/ivmai/bdwgc/commit/be148f4141ab5d6812928a7bceeb2617384e7fe7

[dimpase]

Last 10 new commits:

​871bc6a	7.6.2 is the current version
​71e46a4	On Cygwin install libatomic_ops as a shared lib only; this is inline with what the official Cygwin package does
​0308045	Convert libatomic_ops to use sage-dist-helpers
​8c47a8a	update to 7.6.0
​6148efd	patch for FreeBSD
​7496a68	backported patch for FreeBSD support
​674a6d6	removing already included patches
​48766dd	update gc to 7.6.2
​cdb00d4	For Cygwin only builds shared libraries
​0263add	updated to 7.6.4 and removed patch

[jdemeyer]

1. build/pkgs/libatomic_ops/spkg-install: you are wrongly using "BoehmGC".

2. build/pkgs/libatomic_ops/spkg-install: instead of a generic variable like CONFIGURE_FLAGS, please use LIBATOMIC_OPS_CONFIGURE to be more consistent with other packages.

3. build/pkgs/libatomic_ops/spkg-check: there is no longer need for the if [ -z "$SAGE_LOCAL" ]; then boilerplate.

[dimpase]

Jeroen, libatomic_ops has nothing to do with this ticket. It was #23701. If you think #23701 needs an improvement please open another ticket.

[jdemeyer]

They were just minor comments, nothing serious.

[jdemeyer]

Replying to dimpase:

Jeroen, libatomic_ops has nothing to do with this ticket. It was #23701. If you think #23701 needs an improvement please open another ticket.

To be very precise, the commits which are supposedly from #23701 do not actually appear on the branch at #23701. So this ticket needs to be rebased.

[jdemeyer]

There is also the obvious missing dependency

[gc-7.6.4] checking for atomic_ops... no
[gc-7.6.4] checking atomic_ops.h usability... no
[gc-7.6.4] checking atomic_ops.h presence... no
[gc-7.6.4] checking for atomic_ops.h... no
[gc-7.6.4] configure: error: libatomic_ops is required.  You can either install it on
[gc-7.6.4]                   your system, or fetch and unpack a recent version into the
[gc-7.6.4]                   source directory and link or rename it to libatomic_ops.
[gc-7.6.4] Error configuring BoehmGC.

[dimpase]

Replying to jdemeyer:

There is also the obvious missing dependency

[gc-7.6.4] checking for atomic_ops... no
[gc-7.6.4] checking atomic_ops.h usability... no
[gc-7.6.4] checking atomic_ops.h presence... no
[gc-7.6.4] checking for atomic_ops.h... no
[gc-7.6.4] configure: error: libatomic_ops is required.  You can either install it on
[gc-7.6.4]                   your system, or fetch and unpack a recent version into the
[gc-7.6.4]                   source directory and link or rename it to libatomic_ops.
[gc-7.6.4] Error configuring BoehmGC.

good catch, but how do you get this? OS/compiler?

[jdemeyer]

Replying to dimpase:

OS/compiler?

Doesn't matter. Any OS, any compiler.

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​4d6eee3

added deps; addressed reviewer comments: #23700:37

[dimpase]

all fixed, hopefully.

[jdemeyer]

1. This branch still needs to be rebased to the branch at #23701.

2. You should replace CONFIGURE_FLAGS -> LIBATOMIC_OPS_CONFIGURE everywhere in build/pkgs/libatomic_ops/spkg-install

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​4d5bbd0	libatomic_ops 7.4.6 added as standard pkg
​7b0f626	Merge branch 'u/dimpase/libatomic_ops' of trac.sagemath.org:sage into laup
​5c5d8f6	remove meaningless flags
​54e1da7	Merge branch 'develop' into libatomic_ops
​a024f8d	7.6.2 is the current version
​7135b2f	On Cygwin install libatomic_ops as a shared lib only; this is inline with what the official Cygwin package does
​18813a2	Convert libatomic_ops to use sage-dist-helpers
​8d05b70	Merge branch 'atomic762' into gc764
​cd63b82	LIBATOMIC_OPS_CONFIGURE throughout

[dimpase]

I've addressed the comment 46.

[jdemeyer]

The git history is a total mess now... at least messy enough that I'd like to see it merged with the next beta before considering to review it.

[dimpase]

well, yes, I tried to rebase, and it was really doing weird stuff, such as creating LICENSE file and changing .gitignore, so I merged instead.

[embray]

Perhaps I could try to clean it up?

[dimpase]

Hopefully it is clean enough now, as the merge of #23701 has happened.

[jdemeyer]

Replying to dimpase:

Hopefully it is clean enough now

It's not cleaner than before since the branch didn't change. The fact that #23701 was merged does mean that the diff is easier to read on Trac.

[dimpase]

The diff is pretty small now: there are largish patches removed, and the rest it really "admin".

Erik, does the branch work on Cygwin for you now? (You never answered comment 34 above). The commit in the upstream repo seems to do the same job and is authored by you, but it looks differently---hopefully it does the same thing.

[jdemeyer]

Replying to dimpase:

The diff is pretty small now

Yes, the diff is small but the history remains ugly: there are several commits on this branch which are really duplicates of commits from #23701.

Now, all that doesn't really matter.

[embray]

LGTM. My offer still stands to clean up the commit history if desired, but otherwise I'll let it be.