GCC and binutils plans for bullseye
Debian bullseye will be based on a gcc-10 package taken from the gcc-10 upstream branch, and binutils based on a binutils package taken from the 2.35 branch. I'm planning to make gcc-10 the default after gcc-10 (10.2.0) is available (upstream targets mid July). binutils will be updated before making the GCC switch. The GCC 10 switch involves some minor library transitions for D, gccgo, M2, which should be no-brainers. The gnat transition will be handled separately by the debian Ada maintainers. binutils should be pretty stable until the bullseye release, not planning an update to 2.36. GCC 10 should be updated to 10.3, or close to 10.3 (the release date is not yet known, could be Feb 2021). I'd like to get rid off GCC 8 and GCC 9 for the bullseye release. Matthias
Re: GCC and binutils updates for buster
2018-08-13 04:25 Paul Wise: On Mon, Aug 13, 2018 at 1:19 AM, Manuel A. Fernandez Montecelo wrote: 2018-07-30 22:36 Adrian Bunk: And the next burden will be if riscv64 gets added in bullseye. [*] Unlike other arches, this one is not restricted to a single vendor so hardware can be annouced at any time from unexpected parties; still, only a few months are left. Only a few months are left before buster, but Adrian was talking about bullseye, which is quite a long way off. Right, I quoted that phrase but my mind was with the Buster in the subject. Thanks for the correction, and sorry for the noise from my side. -- Manuel A. Fernandez Montecelo
Re: GCC and binutils updates for buster
On Mon, Aug 13, 2018 at 1:19 AM, Manuel A. Fernandez Montecelo wrote: > 2018-07-30 22:36 Adrian Bunk: >> >> And the next burden will be if riscv64 gets added in bullseye. > > [*] Unlike other arches, this one is not restricted to a single vendor >so hardware can be annouced at any time from unexpected parties; >still, only a few months are left. Only a few months are left before buster, but Adrian was talking about bullseye, which is quite a long way off. -- bye, pabs https://wiki.debian.org/PaulWise
Re: GCC and binutils updates for buster
2018-07-30 22:36 Adrian Bunk: And the next burden will be if riscv64 gets added in bullseye. Not likely, I think, since for example there's almost no hardware available for end-users to buy (or to use for buildds), and this will probably be the case at least until the freeze [*]. Another reason is that there're missing key components that need to get to the main upstream repos, like GDB, LLVM, Rust, JIT support for OpenJDK, etc. GDB is being upstreamed right now, but there's still way to go for the rest. [*] Unlike other arches, this one is not restricted to a single vendor so hardware can be annouced at any time from unexpected parties; still, only a few months are left. -- Manuel A. Fernandez Montecelo
Re: GCC and binutils updates for buster
On Mon, Jul 16, 2018 at 05:59:28PM +0200, Matthias Klose wrote: >... > - armel: The armv4t default isn't used very much anymore, The baseline is armv5te since last year. > and we had issues in the past. Could you elaborate on that? The latest major issue I am aware of was about #727621 and the backport of the fix.[1] #727621 was not even a regression, this was new functionality accidentally not provided by gcc on armel. > - armhf: While arm-linux-gnueabihf is not explicitly listed as a primary >architecture, I'm told that the arm-linux-armeabi triplet covers the >hard float variants as well. > > - ppc64el: Not documented as primary architecture, but according to the >backend maintainers the powerpc64-linux-gnu triplet includes the le > variant. > > - mips*: There is no support for any mips-linux target either as a primary >or secondary release architecture (only bare metal), which matches the >experience with mips specific issues for the past Debian releases. > > I understand that port maintainers want to have their port included as a > release > architecture, however it becomes a burden if neither the upstream nor the > Debian > port maintainers can keep up with the general upstream development. Maybe we > need something in between the alternatives of being a release arch or not, > having the benefit of packages in testing/stable, but not being supported in a > release. Your theoretical discussion based on upstream definitions misses which architectures actually have a proven track record of frequent toolchain problems in recent years. Any discussion about real problems has to include arm64 as a prime candidate for demotion - no matter the upstream definition. We even had last-minute toolchain regressions like #863152, which means that we cannot rebuild some packages we ship for arm64 in stretch. Such regressions are a problem both for security support and licence reasons. The root problem for most actual breakages is that regresssions are usually not in boring stale old code nobody touches - regressions tend to be in sexy new code that is under heavy development. arm64 and mips64el are the most recent of the architectures in stretch. There is a lot of upstream toolchain development happening for arm64. These are the reasons why arm64 and mips64el have been a burden in recent years. And the next burden will be if riscv64 gets added in bullseye. I do not have the impression that this burden is unmanageable, but if you disagree the actual discussion you have to start is about delaying the inclusion of ports for new hardware in a Debian stable release. > Matthias >... cu Adrian [1] Reminds me that I have to check which Breaks are missing for that. -- "Is there not promise of rain?" Ling Tan asked suddenly out of the darkness. There had been need of rain for many days. "Only a promise," Lao Er said. Pearl S. Buck - Dragon Seed
Re: GCC and binutils updates for buster
On 2018-07-16 17:59:28 [+0200], Matthias Klose wrote: > architectures. Some notes on other candidates for release architectures: > > - armel: The armv4t default isn't used very much anymore, and we had >issues in the past. Would things get better with armv5te as default or is the lack of FPU the actual problem? > Matthias Sebastian
GCC and binutils updates for buster
GCC 8 is available in testing/unstable, and upstream is approaching the first point release. I am planning to make GCC 8 the default at the end of the week (gdc and gccgo already point to GCC 8). Most runtime libraries built from GCC are already used in the version built from GCC 8, so I don't expect runtime incompatibilities anymore. There is one more transistion involved, bumping the libgfortran version. A pre-release version of binutils 2.31 is in testing now, and the final 2.31 release in unstable. These are the major versions for the upcoming buster release, still planning updates to a potential GCC 8.3.0 (estimated Jan 2019) release and binutils 2.31.1 release, or doing equivalent updates from the release branches. There are still a bunch of build failures triggered by GCC 8 [1], so fixing these should get some priority now. See [2] for changes in GCC 8, and the porting guide [3]. I'll be at DebCamp for the second half of the week, and at DebConf, so if there is interest for bug squashing sessions, feel free to grab me, and we can schedule such sessions on a short notice. GCC 5 and GCC 6 are going away, and I am planning the same with GCC 7 as soon as there are upstream kernel and glibc releases which are released after the GCC 8.1.0 release from April. The Debian release team lists toolchain support for our release architectures, and according to [4], the amd64, i386, armel, armhf, arm64 architectures are supported as primary architectures, and s390x is supported as a secondary architectures. Some notes on other candidates for release architectures: - armel: The armv4t default isn't used very much anymore, and we had issues in the past. - armhf: While arm-linux-gnueabihf is not explicitly listed as a primary architecture, I'm told that the arm-linux-armeabi triplet covers the hard float variants as well. - ppc64el: Not documented as primary architecture, but according to the backend maintainers the powerpc64-linux-gnu triplet includes the le variant. - mips*: There is no support for any mips-linux target either as a primary or secondary release architecture (only bare metal), which matches the experience with mips specific issues for the past Debian releases. I understand that port maintainers want to have their port included as a release architecture, however it becomes a burden if neither the upstream nor the Debian port maintainers can keep up with the general upstream development. Maybe we need something in between the alternatives of being a release arch or not, having the benefit of packages in testing/stable, but not being supported in a release. Matthias [1] http://bugs.debian.org/cgi-bin/pkgreport.cgi?tag=ftbfs-gcc-8;users=debian-...@lists.debian.org [2] https://gcc.gnu.org/gcc-8/changes.html [3] https://gcc.gnu.org/gcc-8/porting_to.html [4] https://gcc.gnu.org/gcc-8/criteria.html signature.asc Description: OpenPGP digital signature
Re: Reply headers vs. in-mail requests (was Re: Bad news to CUDA applications (was: Re: GCC 6 & binutils for the Debian stretch release))
On Fri, Jul 01, 2016 at 09:04:50AM -0400, The Wanderer wrote: > I'm pretty sure there is no header setting which will make this fully > automatic. I have been suitably admonished off-list for starting this $DEITY-awful sub-topic once again. Consider myself chastened! -- Jonathan Dowland Please do not CC me, I am subscribed to the list. signature.asc Description: Digital signature
Reply headers vs. in-mail requests (was Re: Bad news to CUDA applications (was: Re: GCC 6 & binutils for the Debian stretch release))
On 2016-07-01 at 08:43, Jonathan Dowland wrote: > On Fri, Jul 01, 2016 at 06:07:27AM +, lumin wrote: > >> (please keep me in CC list) > > I suggest setting the headers to make this automatic rather than > asking people to remember to do it :) I'm pretty sure there is no header setting which will make this fully automatic. For one thing, he can only control header settings on mails which he sends, not on mails which other people send in reply to those mails. If he wants to be CCed on replies to those replies, as far as I know there is no header which he can set which will make that happen. For another thing, some replying practices will ignore such headers. I normally reply to list mail using Thunderbird/Icedove's "Reply to List" button, which addresses the reply only to the mailing list itself, ignoring any headers which may be set to direct replies elsewhere. I do this because otherwise the reply tends to be directed either to the original sender rather than to the list (with "Reply", in the absence of Reply-To set to the list address), or to all recipients (with "Reply All"), depending on what headers are present - and IMO neither of those is the correct default. The presence of a comment such as this one informs me that I may want to vary that practice in a given case. (I haven't done so in this case since this mail is no longer on the topic which he raised, so I don't have good reason to expect him to be interested in it.) -- The Wanderer The reasonable man adapts himself to the world; the unreasonable one persists in trying to adapt the world to himself. Therefore all progress depends on the unreasonable man. -- George Bernard Shaw signature.asc Description: OpenPGP digital signature
Re: Bad news to CUDA applications (was: Re: GCC 6 & binutils for the Debian stretch release)
On Fri, Jul 01, 2016 at 06:07:27AM +, lumin wrote: > (please keep me in CC list) I suggest setting the headers to make this automatic rather than asking people to remember to do it :) > I'm pointing out a BIG problem introduced by stretch's GCC-6-only plan. > > In brief CUDA 8.0~RC fails to work with GCC-6, this conclusion > comes from my local Caffe build log as attached. > That is to say, after GCC-6 transition *ALL* packages depending > on cuda will get removed from stretch due to FTBFS. Technically, these packages are/were never in Stretch per se, because they're all contrib/non-free. -- Jonathan Dowland Please do not CC me, I am subscribed to the list. signature.asc Description: Digital signature
Re: Bad news to CUDA applications (was: Re: GCC 6 & binutils for the Debian stretch release)
Le 01/07/2016 à 08:51, lumin a écrit : > > Releated bug on ArchLinux: > https://bugs.archlinux.org/task/49272?project=5=12602 > > There are some hacks but none of them seems to be "an actual solution > to packaging". Personally, I would create a gcc/g++ wrapper in order to capture the exact options passed by nvcc and to record the input file(s). Then, I would try to see how things can be fixed manually (ie find options to add to the real gcc call to compile successfully). And, if nvidia does not update nvcc (and if the previous step was successful), I would create a wrapper around nvcc that force to call a gcc-wrapper (perhaps just redefining PATH would be enough) that would add fix-flags to a real gcc call. Regards, Vincent > On Fri, 2016-07-01 at 06:07 +, lumin wrote: >> Hi all, >> (please keep me in CC list) >> >> I'm pointing out a BIG problem introduced by stretch's GCC-6-only plan. >> >> In brief CUDA 8.0~RC fails to work with GCC-6, this conclusion >> comes from my local Caffe build log as attached. >> That is to say, after GCC-6 transition *ALL* packages depending >> on cuda will get removed from stretch due to FTBFS. >> >> I don't expect Nvidia to release CUDA 8.5 before the stretch >> freeze date (Q1 2017), i.e. even a freeze-exception against >> cuda might not save this situation. So all maintainers maintaining >> CUDA application packages have to face this harsh condition. >> Do you have any solution to this problem? >> >> Besides, I cc'ed 2 nvidia guys with a hope that they can provide >> some helpful information. > > -- Vincent Danjean GPG key ID 0xD17897FA vdanj...@debian.org GPG key fingerprint: 621E 3509 654D D77C 43F5 CA4A F6AE F2AF D178 97FA Unofficial pkgs: http://moais.imag.fr/membres/vincent.danjean/deb.html APT repo: deb http://people.debian.org/~vdanjean/debian unstable main
Re: Bad news to CUDA applications (was: Re: GCC 6 & binutils for the Debian stretch release)
Releated bug on ArchLinux: https://bugs.archlinux.org/task/49272?project=5=12602 There are some hacks but none of them seems to be "an actual solution to packaging". On Fri, 2016-07-01 at 06:07 +, lumin wrote: > Hi all, > (please keep me in CC list) > > I'm pointing out a BIG problem introduced by stretch's GCC-6-only plan. > > In brief CUDA 8.0~RC fails to work with GCC-6, this conclusion > comes from my local Caffe build log as attached. > That is to say, after GCC-6 transition *ALL* packages depending > on cuda will get removed from stretch due to FTBFS. > > I don't expect Nvidia to release CUDA 8.5 before the stretch > freeze date (Q1 2017), i.e. even a freeze-exception against > cuda might not save this situation. So all maintainers maintaining > CUDA application packages have to face this harsh condition. > Do you have any solution to this problem? > > Besides, I cc'ed 2 nvidia guys with a hope that they can provide > some helpful information.
Bad news to CUDA applications (was: Re: GCC 6 & binutils for the Debian stretch release)
Hi all, (please keep me in CC list) I'm pointing out a BIG problem introduced by stretch's GCC-6-only plan. In brief CUDA 8.0~RC fails to work with GCC-6, this conclusion comes from my local Caffe build log as attached. That is to say, after GCC-6 transition *ALL* packages depending on cuda will get removed from stretch due to FTBFS. I don't expect Nvidia to release CUDA 8.5 before the stretch freeze date (Q1 2017), i.e. even a freeze-exception against cuda might not save this situation. So all maintainers maintaining CUDA application packages have to face this harsh condition. Do you have any solution to this problem? Besides, I cc'ed 2 nvidia guys with a hope that they can provide some helpful information. caffe_buildlog_nvcc8_gcc6_failure.txt.gz Description: application/gzip signature.asc Description: This is a digitally signed message part
Re: GCC 6 & binutils for the Debian stretch release
On Fri, Jun 24, 2016 at 04:03:44PM +0200, Paul Wise wrote: > On Fri, Jun 24, 2016 at 3:46 PM, Matthias Klose wrote: > > > As announced a year ago [1], GCC 6 will be the default GCC for the Debian > > stretch release. GCC 6 is now available in testing, and can be made the > > default > > by installing the gcc/g++ packages from experimental. Known build failures > > are > > reported at [2] seen on amd64. Build failures for more architectures (but > > done > > for Ubuntu packages) can be seen at [3]. Please help fixing these issues in > > testing/unstable. Some help how to approach build issues can be found at > > [4]. > > Could we have a dd-list of people who will have to fix a bug for this > transition? Attached, based on “bts select users:debian-...@lists.debian.org tag:ftbfs-gcc-6 status:open”. Cheers, -- James GPG Key: 4096R/91BF BF4D 6956 BD5D F7B7 2D23 DFE6 91AE 331B A3DB "Adam C. Powell, IV"elmerfem (U) oce (U) "Natural Language Processing, Japanese" chasen A. Maitland Bottoms sdrangelove splat (U) Adrian Knoth calf (U) jackd2 (U) libdrumstick (U) Agustin Henze crrcsim Aigars Mahinovs re Alan Baghumian dasher (U) Alastair McKinstry ggcov Alessio Treglia ams (U) composite (U) crtmpserver (U) din (U) klick (U) libdrumstick (U) sndobj (U) terminatorx (U) Alexander GQ Gerasiov gxneur xneur Alexey Bychko percona-xtrabackup (U) Ana Beatriz Guerrero Lopez mstflint (U) Andreas Tille adun.app (U) blitz++ (U) disulfinder (U) hyphy (U) idba (U) libbpp-core (U) libgtextutils (U) libmems (U) liborigin2 (U) librg-blast-parser-perl (U) librostlab-blast (U) mrs (U) murasaki (U) prime-phylo (U) proftmb (U) qwtplot3d (U) rate4site (U) relion (U) sitplus (U) sofa-framework (U) swarm-cluster (U) yaml-cpp (U) Andres Mejia crtmpserver (U) Andrew Bartlett samba (U) Andrew Pollock protobuf (U) Andrew Shadura postbooks (U) Andriy Beregovenko crtmpserver (U) Ansgar Burchardt dune-grid (U) Anton Gladky avogadro (U) esys-particle (U) oce (U) paraview (U) Anuradha Weeraman (anu) ncc Apollon Oikonomopoulos mongodb (U) Arnout Engelen libdrumstick (U) Aron Xu fcitx-unikey (U) librime (U) libucimf (U) Aron Xu ucimf-openvanilla (U) Athena Capital Research pion Balint Reczey dasher (U) libcec-platform (U) libv8-3.14 (U) Barak A. Pearlmutter colpack (U) ivtools mldemos Barry deFreese bloboats (U) Barry deFreese libclaw (U) Bastian Blank thin-provisioning-tools (U) xen (U) Bdale Garbee splat Ben Burton regina-normal Ben Hutchings sgt-puzzles Benda Xu scim (U) Benjamin Drung audacity (U) Bernd Zeimetz open-vm-tools Boris Pek elmerfem (U) Bradley Smith galib Bruno "Fuddl" Kleinert scorched3d (U) Bryan Sutula openhpi Carlo Segre fityk objcryst-fox Ceph Maintainers ceph ChangZhuo Chen (陳昌倬) libucimf (U) Charles Plessy libgtextutils (U) Chow Loong Jin libzen slic3r (U) tinyxml2 Chris Butler libcgicc Chris Coulson mozjs Christian M. Amsüss hyperrogue (U) Christian Perrier samba (U) Christian Stalp free42-nologo Christoph Egger fife (U) irrlicht (U) Christophe Prud'homme paraview (U) Christophe Trophime blitz++ (U) Cleto Martín zthreads Craig Small mudlet CrossWire Packages sword Cédric Boutillier gfan (U) Daigo Moriwaki gpsshogi libosl Damyan Ivanov gbgoffice hyperrogue (U) Daniel Glassey clucene-core
Re: BOOST-1.60 compiled with g++6, [Was: GCC 6 & binutils for the Debian stretch release]
Hello, On 27 June 2016 at 11:37, Dimitri John Ledkovwrote: > Hello, > > On 26 June 2016 at 11:31, Gert Wollny wrote: >> Hi all, >> >> considering that BOOST 1.60 changes the ABI when compiled with -std >= >> c++11 versus -std <= c++03 (cf. [1]) , and that g++-6 defaults to >> -std=c++14 it would probably be a good idea if a boost >= 1.60 version >> compiled with g++6 would be available from experimental when the bug >> squashing starts. > > gnu/c++03 -> gnu/c++11 was special as there was libstdc++ library ABI break. > > As far as I am aware gcc6 changes the default _standard_ version to > gnu++14, however it remains binary compatible with gnu/c++11 libstdc++ > ABI. > > In other words, please do not assume that we need to bump ABI across > the board the way we did when migrating to libstdc++ 11 abi. > > I will do more checking, but my understanding is that current boost > should be fit for purpose for either gnu++11 or gnu++14 standard > builds with c++11 libstdc++ ABI. Debian no longer supports libstdc++ > c++03 abi. However above seems to contradict my understanding of the situation in May 2016 lol =) Back then I did believe that boost needs to be rebuild. *sigh* i'm really not in a rational state of mind at the moment. I'll try to dig into it during Debconf 2016, when I'm away from the current politics in my country. -- Regards, Dimitri.
Re: BOOST-1.60 compiled with g++6, [Was: GCC 6 & binutils for the Debian stretch release]
Hello, On 26 June 2016 at 11:31, Gert Wollnywrote: > Hi all, > > considering that BOOST 1.60 changes the ABI when compiled with -std >= > c++11 versus -std <= c++03 (cf. [1]) , and that g++-6 defaults to > -std=c++14 it would probably be a good idea if a boost >= 1.60 version > compiled with g++6 would be available from experimental when the bug > squashing starts. gnu/c++03 -> gnu/c++11 was special as there was libstdc++ library ABI break. As far as I am aware gcc6 changes the default _standard_ version to gnu++14, however it remains binary compatible with gnu/c++11 libstdc++ ABI. In other words, please do not assume that we need to bump ABI across the board the way we did when migrating to libstdc++ 11 abi. I will do more checking, but my understanding is that current boost should be fit for purpose for either gnu++11 or gnu++14 standard builds with c++11 libstdc++ ABI. Debian no longer supports libstdc++ c++03 abi. -- Regards, Dimitri.
BOOST-1.60 compiled with g++6, [Was: GCC 6 & binutils for the Debian stretch release]
Hi all, considering that BOOST 1.60 changes the ABI when compiled with -std >= c++11 versus -std <= c++03 (cf. [1]) , and that g++-6 defaults to -std=c++14 it would probably be a good idea if a boost >= 1.60 version compiled with g++6 would be available from experimental when the bug squashing starts. Best, Gert [1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=823978
Re: GCC 6 & binutils for the Debian stretch release
On Fri, Jun 24, 2016 at 3:46 PM, Matthias Klose wrote: > As announced a year ago [1], GCC 6 will be the default GCC for the Debian > stretch release. GCC 6 is now available in testing, and can be made the > default > by installing the gcc/g++ packages from experimental. Known build failures > are > reported at [2] seen on amd64. Build failures for more architectures (but > done > for Ubuntu packages) can be seen at [3]. Please help fixing these issues in > testing/unstable. Some help how to approach build issues can be found at [4]. Could we have a dd-list of people who will have to fix a bug for this transition? -- bye, pabs https://wiki.debian.org/PaulWise
Re: gcc and binutils
Bernd Eckenfels writes (gcc and binutils): ist it possile that on a fresh new install gcc is installed before binutils is installed, and therefore fail to configure? If I run configure afterwards everything is fine. Will dpkg install a package first if it sees that other ones depend on it? Err, I don't quite understand the question. The answer to your first sentence is `no, it shouldn't be possible'. See the draft programmers' manual for details of exactly how things happen. Ian.
gcc and binutils
Hi, ist it possile that on a fresh new install gcc is installed before binutils is installed, and therefore fail to configure? If I run configure afterwards everything is fine. Will dpkg install a package first if it sees that other ones depend on it? Greetings Bernd -- (OO) -- [EMAIL PROTECTED] -- ( .. ) [EMAIL PROTECTED],ka.sub.org} http://home.pages.de/~eckes/ o--o *plush* 2048/A2C51749 [EMAIL PROTECTED] +4972573817 *plush* (OO) If privacy is outlawed only Outlaws have privacy