Opened 2 years ago
Closed 11 months ago
#65188 closed defect (fixed)
sbcl fails to build on PPC: fatal error encountered in SBCL pid 16153: dynamic space too small for core
| Reported by: | barracuda156 | Owned by: | easye |
|---|---|---|---|
| Priority: | Normal | Milestone: | |
| Component: | ports | Version: | 2.7.2 |
| Keywords: | powerpc | Cc: | kakuhen, acjones8 (Alex Jones), amock (Alan Mock) |
| Port: | sbcl |
Description
I am trying to get sbcl to build for PPC. This ticket is for @2.2.2, which is not the current version in Macports. However please bear with me here. I am pretty sure restoring PPC support into @2.2.4 is relatively trivial once @2.2.2 is confirmed to build (it is the last version having PPC code before it was removed). (In fact I have preliminary patches for @2.2.4.)
I need to get past the following error. I initially got it when using @1.2.8 to build @2.2.2. Then I built @1.5.9 with @1.2.8 and tried again, same failure.
//entering make-target-2.sh //doing warm init - compilation phase This is SBCL 2.2.2, an implementation of ANSI Common Lisp. More information about SBCL is available at <http://www.sbcl.org/>. SBCL is free software, provided as is, with absolutely no warranty. It is mostly in the public domain; some portions are provided under BSD-style licenses. See the CREDITS and COPYING files in the distribution for more information. fatal error encountered in SBCL pid 16153: dynamic space too small for core: 31044KiB required, 0KiB available. Command failed: SBCL_MACOSX_VERSION_MIN=10.6 CC=/usr/bin/gcc-4.2 CXX=/usr/bin/g++-4.2 CPP=/usr/bin/cpp-4.2 sh ./make.sh --prefix=/opt/local --xc-host="/opt/local/share/sbcl-ppc/bin/sbcl --core /opt/local/share/sbcl-ppc/lib/sbcl/sbcl.core --disable-debugger --sysinit /dev/null --userinit /dev/null" --dynamic-space-size=8Gb Exit code: 1
I added --dynamic-space-size=8Gb following some suggestion online, it did not help. (At first I did not specify this flag.)
- S. A caveat: threads are broken (might be fixable), so my builds are without variants added.
Attachments (4)
Change History (41)
comment:1 Changed 2 years ago by barracuda156
comment:2 Changed 2 years ago by barracuda156
Just found this: https://github.com/macports/macports-ports/commit/e5b44a273814eb5a05f780bf9fc22f94d0f487a8 Will try back-porting to @1.5.9 now.
comment:3 Changed 2 years ago by barracuda156
Some test failures with 1.2.8 are not PPC-specific: https://sourceforge.net/p/sbcl/mailman/message/21453445/
comment:4 follow-up: 5 Changed 2 years ago by easye
@barracuda156 Thanks for the hard work here. Of course I wish I had access to a PPC system to help, but perhaps I can coordinate more fully with the SBCL crew on IRC <irc://libera.chat/#sbcl> or the SBCL bug tracker <https://launchpad.net/sbcl> if you need something? Feel free to contact me "out of band".
comment:5 follow-up: 21 Changed 2 years ago by barracuda156
Replying to easye:
@barracuda156 Thanks for the hard work here. Of course I wish I had access to a PPC system to help, but perhaps I can coordinate more fully with the SBCL crew on IRC <irc://libera.chat/#sbcl> or the SBCL bug tracker <https://launchpad.net/sbcl> if you need something? Feel free to contact me "out of band".
Thank you for replying! There are two immediate problems I face, and any help will be greatly appreciated here:
- There is a section in
src/runtime/ppc-assem.Swhich has never been fixed for Darwin ABI:/* Call out to obtain our TLS block. */ load(reg_NL0,CSYMBOL(specials)) /* This won't work on darwin: wrong fixup style. And is it * supposed to be lis/ori or lis/addi? Or does it differ * between darwin and everything else again? */ lis reg_CFUNC,CSYMBOL(pthread_getspecific)@h ori reg_CFUNC,reg_CFUNC,CSYMBOL(pthread_getspecific)@l mtctr reg_CFUNC bctrl mr reg_THREAD, reg_NL0
Those suffices @ are invalid, of course, and the code fails. However it is required in order to build with threads enabled. I tried patching it to:
+#ifdef LISP_FEATURE_DARWIN + lis reg_CFUNC,hi16(CSYMBOL(pthread_getspecific)) + ori reg_CFUNC,reg_CFUNC,lo16(CSYMBOL(pthread_getspecific)) +#else lis reg_CFUNC,CSYMBOL(pthread_getspecific)@h ori reg_CFUNC,reg_CFUNC,CSYMBOL(pthread_getspecific)@l +#endif
This fails however:
ld: absolute address to symbol _pthread_getspecific in a different linkage unit not supported in _call_into_lisp from ppc-assem.o collect2: ld returned 1 exit status make: *** [sbcl] Error 1
This is why I had to build without threads (well, I don't promise threads will work if this is fixed, but at least this must be fixed in some way in order to proceed). It has been broken forever: from @1.0.47 up to @2.2.2 this chunk of code remains unchanged, and in @2.2.3 upstream, sadly, just threw away Darwin-specific parts. (So I think it has never been built for PPC at all.)
- The issue in the ticket above:
fatal error encountered in SBCL pid 16153: dynamic space too small for core: 31044KiB required, 0KiB available.
Since it is not pointing to a specific error in the code, I have no hint how to try fixing it.
comment:6 follow-ups: 7 24 Changed 2 years ago by easye
Alright. AFK at the moment (and broadly "working" in CEST as a time zone, so will see what info I can dig up in the next 12 hours.
comment:7 follow-up: 8 Changed 2 years ago by barracuda156
Replying to easye:
Alright. AFK at the moment (and broadly "working" in CEST as a time zone, so will see what info I can dig up in the next 12 hours.
Great! Thank you very much!
comment:8 Changed 2 years ago by easye
Replying to barracuda156: Unfortunately no progress to report. Getting in touch with SBCL maintainers may take some time.
More when I know…
comment:9 follow-ups: 10 11 25 Changed 2 years ago by easye
Support for Darwin/PPC was removed with sbcl-2.2.3 <http://sbcl.org/news.html>.
Gonna be kinda tricky to do this for a single Portfile, so I wonder if anyone can chime in on what should be done: a "new" port with the architecture in the name ('lang/sbcl-ppc')?
comment:10 follow-ups: 12 27 Changed 2 years ago by easye
Replying to easye:
Support for Darwin/PPC was removed with sbcl-2.2.3 <http://sbcl.org/news.html>.
A statement from the SBCL maintainer via IRC:
<Krystof> easye: I think it's worth saying that if there are users willing to
step up and maintain our ppc/darwin support, we'd reinstate it on
that basis [13:20]
<Krystof> the problem we were having is: no hardware, no access to hardware,
no demand from users, no ability to check whether changes for other
platforms cause breakage [13:21]
comment:11 Changed 2 years ago by barracuda156
Replying to easye:
Support for Darwin/PPC was removed with sbcl-2.2.3 <http://sbcl.org/news.html>.
Gonna be kinda tricky to do this for a single Portfile, so I wonder if anyone can chime in on what should be done: a "new" port with the architecture in the name ('lang/sbcl-ppc')?
What I had already done is sbcl-ppc-bootstrap port that uses 1.0.47 (the last available binary) to build 1.2.8 (maximum it can build, same for Intel), and then I used 1.2.8 to build 1.5.9. Versions 2.x.x do not build yet, however.
Yes, I am aware that upstream removed PPC code in 2.2.3, that’s why I refer to 2.2.2 in the ticket. However it seems not too hard to revert the removal.
comment:12 follow-up: 13 Changed 2 years ago by barracuda156
Replying to easye:
Replying to easye:
Support for Darwin/PPC was removed with sbcl-2.2.3 <http://sbcl.org/news.html>.
A statement from the SBCL maintainer via IRC:
<Krystof> easye: I think it's worth saying that if there are users willing to step up and maintain our ppc/darwin support, we'd reinstate it on that basis [13:20] <Krystof> the problem we were having is: no hardware, no access to hardware, no demand from users, no ability to check whether changes for other platforms cause breakage [13:21]
If the issue is purely technical, then it is not too hard to build PPC code on a modern Intel Mac. Set up 10.6.x in VM, make sure to install Rosetta, build Macports for PPC natively, run PPC binaries. I have such set up on my MacMini at home with Catalina.
But of course, the problem may rather be economical: upstream has limited time and may prefer use it for smth else. This is understandable, so my questions here do not imply I expect others to fix the thing and keep maintaining. It just might happen that for someone from developers errors I reported are known and trivial, while for me it may take days or weeks.
My general understanding (which can be wrong) is that compilers, if properly designed, are cross-platform, and we don’t need to change PPC assembler, it just has to be fixed once. If gcc11 builds, sbcl should ))
- S. I don’t want to create a burden for anyone, and lack of hardware or spare time is perfectly understandable. If someone comes up with a useful suggestion, great, otherwise I will try to fix it myself and update on success or failure.
I will look into assembler error, it’s two lines of code after all, should be fixable.
comment:13 follow-up: 14 Changed 2 years ago by easye
Replying to barracuda156:
Support for Darwin/PPC was removed with sbcl-2.2.3 <http://sbcl.org/news.html>.
A statement from the SBCL maintainer via IRC:
<Krystof> easye: I think it's worth saying that if there are users willing to step up and maintain our ppc/darwin support, we'd reinstate it on that basis [13:20] <Krystof> the problem we were having is: no hardware, no access to hardware, no demand from users, no ability to check whether changes for other platforms cause breakage [13:21]If the issue is purely technical, then it is not too hard to build PPC code on a modern Intel Mac.
[…]
I might be leery of not understanding how Rosetta interacts with my debugging. But maybe I should give this a whirl on the 2013 iMac I have running Catalina.
But of course, the problem may rather be economical: upstream has limited time
[…] SBCL development doesn't have tons of resources. Krystof seems to be implying that they don't really have a demand from users.
Once you get something running in MacPorts we can perhaps assess the effort required to merge the patches upstream. Since they don't seem to have test coverage under ppc/darwin, figuring out how to give them a CI instance they can run to see when they break ppc/darwin would probably win over bonus points.
comment:14 Changed 2 years ago by barracuda156
Replying to easye:
I might be leery of not understanding how Rosetta interacts with my debugging. But maybe I should give this a whirl on the 2013 iMac I have running Catalina.
Please do not try it now, it won't work for sbcl-ppc, at least right-away. The problem is that we are forced to start with pre-built binary, and it fails, confusing the arch with x86_64.
I have the same problem with other ports that rely on pre-built boot binary: my ghc-ppc-bootstrap works on 10.5.8 and 10.6 PPC, but fails on Rosetta, and the recent openjdk7 attempt also fails.
I tried now in order to let you know if it works. I get these funny errors:
CORRUPTION WARNING in SBCL pid 40193: Memory fault at ebebebeb (pc=0x0, sp=0x0) The integrity of this image is possibly compromised.
I tried to trick make-config.sh, forcing ppc, but it did not help.
So yeah, while 90% or more of ports build for PPC on Rosetta, few fail, and this is such a case.
SBCL development doesn't have tons of resources. Krystof seems to be implying that they don't really have a demand from users.
Well, Mac community have to blame themselves, honestly speaking, since they generally keep away from developers, and then developers get impression that no one uses PPC Macs for years. At the same time even more obscure archs like MIPS and SPARC remain supported. Like, how many users of SGI are there? (I wish I had Tezro, but it's just crazy expensive.)
Once you get something running in MacPorts we can perhaps assess the effort required to merge the patches upstream. Since they don't seem to have test coverage under ppc/darwin, figuring out how to give them a CI instance they can run to see when they break ppc/darwin would probably win over bonus points.
In principle, I can arrange remote access to my G5 Quad, if someone would really want to put some effort into this, but as I said, my hopes are very modest.
Changed 2 years ago by barracuda156
| Attachment: | sbcl-ppc-bootstrap.zip added |
|---|
This should build 1.2.8 on Leopard or Snow Leopard PPC (Tiger and Rosetta fail). WIP version: I did a quick cleanup now, hopefully didn't delete anything necessary.
Changed 2 years ago by barracuda156
| Attachment: | sbcl-ppc.zip added |
|---|
Then this should build sbcl 1.5.9, using sbcl-ppc-bootstrap (tested on 10.6 PPC). Variants won't build. WIP, no clean-up
comment:15 follow-ups: 16 17 Changed 2 years ago by tomio-arisaka (Tomio Arisaka)
Thanks for your great efforts!
I'm interested in SBCL. So I tried to install SBCL on PPC Tiger.
I modified the SBCL portfile based on the old one which is got from GitHub.
Then I was able to install SBCL-2.0.9. But I cannot install the later versions than 2.0.9.
("a_sbcl" variant means that the active SBCL is used to compile this version)
$ port installed sbcl The following ports are currently installed: sbcl @1.1.6_1 sbcl @1.2.11_0+a_sbcl sbcl @1.4.9_0+a_sbcl sbcl @1.5.9_0+a_sbcl sbcl @2.0.9_0+a_sbcl (active) $ $ file `which sbcl` /opt/local/bin/sbcl: Mach-O executable ppc $ $ cat << EOF | sbcl --noinform --no-sysinit --no-userinit > (machine-type) > (machine-version) > (software-type) > (software-version) > (lisp-implementation-type) > (lisp-implementation-version) > *features* > *read-default-float-format* > (+ 1 2 3 4 5 6 7 8 9 10) > (expt 2.0 10) > EOF * "PowerPC" * "PowerBook3,5" * "Darwin" * "8.11.0" * "SBCL" * "2.0.9" * (:PPC :GENCGC :ANSI-CL :BIG-ENDIAN :BSD :COMMON-LISP :DARWIN :IEEE-FLOATING-POINT :MACH-O :PACKAGE-LOCAL-NICKNAMES :SB-LDB :SB-PACKAGE-LOCKS :SB-UNICODE :SBCL :UNIX) * SINGLE-FLOAT * 55 * 1024.0 * $
comment:16 follow-up: 19 Changed 2 years ago by barracuda156
Replying to tomio-arisaka:
Thanks for your great efforts!
I'm interested in SBCL. So I tried to install SBCL on PPC Tiger. I modified the SBCL portfile based on the old one which is got from GitHub.
Then I was able to install SBCL-2.0.9. But I cannot install the later versions than 2.0.9.
Awesome!
Could you describe the process step-wise? For me 1.5.9 failed on Tiger (on Leopard and SL PPC it builds fine).
Did you make any custom patches for these?
- S. I did not yet try building 2.0.9. 2.0.11 failed to build. I will try now with 2.0.9.
comment:17 Changed 2 years ago by barracuda156
Replying to tomio-arisaka:
Maybe you can suggest a better procedure than I have used in my portfiles above. Ideally we would want to minimize steps needed to get to the latest (or the latest feasible for PPC) version of sbcl. (In any case we will likely need a new port to bootstrap some working version of sbcl – what I have as sbcl-ppc-bootstrap or alike.)
comment:18 Changed 2 years ago by barracuda156
I confirm that sbcl 2.0.9 builds. In fact I built it with 1.2.8, which means no in-between steps: sbcl-ppc-bootstrap (1.2.8) → sbcl (2.0.9).
36-111% port -v installed | grep sbcl sbcl @1.5.9_0 requested_variants='' platform='darwin 10' archs='ppc' date='2022-05-16T17:16:23+0800' sbcl @2.0.9_0 (active) requested_variants='' platform='darwin 10' archs='ppc' date='2022-05-20T05:22:37+0800' sbcl-ppc @1.5.9_0 (active) requested_variants='' platform='darwin 10' archs='ppc' date='2022-05-16T17:56:36+0800' sbcl-ppc-bootstrap @1.2.8_0 (active) requested_variants='' platform='darwin 10' archs='ppc' date='2022-05-16T11:18:46+0800'
Changed 2 years ago by tomio-arisaka (Tomio Arisaka)
| Attachment: | MacPorts-SBCL-PPC-Tiger.tar.bz2 added |
|---|
old SBCL portfiles for PPC Tiger
comment:19 follow-up: 20 Changed 2 years ago by tomio-arisaka (Tomio Arisaka)
Replying to barracuda156:
Replying to tomio-arisaka:
Thanks for your great efforts!
I'm interested in SBCL. So I tried to install SBCL on PPC Tiger. I modified the SBCL portfile based on the old one which is got from GitHub.
Then I was able to install SBCL-2.0.9. But I cannot install the later versions than 2.0.9.
Awesome!
Could you describe the process step-wise? For me 1.5.9 failed on Tiger (on Leopard and SL PPC it builds fine).
Did you make any custom patches for these?
- S. I did not yet try building 2.0.9. 2.0.11 failed to build. I will try now with 2.0.9.
"MacPorts-SBCL-PPC-Tiger.tar.bz2" file includes the old SBCL portfiles modified by me. I think they work, but they are not good scripts:)
The sequence of installing SBCL-2.0.9 is as follows:
$ port info sbcl sbcl @1.1.6_1 (lang) Variants: fancy, html, pdf, threads ... ... $ sudo port install sbcl $ port installed The following ports are currently installed: apple-gcc42 @5666.3_16+bootstrap (active) gcc_select @0.1_9 (active) sbcl @1.1.6_1 (active) $ sudo port install apple-gcc42 $ port installed The following ports are currently installed: apple-gcc42 @5666.3_16+bootstrap apple-gcc42 @5666.3_16 (active) cctools @949.0.1_2 (active) gcc_select @0.1_9 (active) gettext @0.21_0 (active) gettext-runtime @0.21_0 (active) gettext-tools-libs @0.21_0 (active) gperf @3.1_0 (active) ld64 @3_4+ld64_97 (active) ld64-97 @97.17_9 (active) libiconv @1.16_1 (active) libmacho-headers @949.0.1_0 (active) libtextstyle @0.21_0 (active) libunwind-headers @5.0.1_0 (active) ncurses @6.3_0 (active) sbcl @1.1.6_1 (active) xz @5.2.5_1 (active) $ port info sbcl sbcl @1.2.11 (lang) Variants: a_sbcl, [+]fancy, html, pdf, threads ... ... $ sudo port install sbcl +a_sbcl -fancy $ port installed sbcl The following ports are currently installed: sbcl @1.1.6_1 sbcl @1.2.11_0+a_sbcl (active) $ port info sbcl sbcl @1.4.9 (lang) Variants: a_sbcl, [+]fancy, html, pdf, threads ... ... $ sudo port install sbcl -fancy +a_sbcl $ port installed sbcl The following ports are currently installed: sbcl @1.1.6_1 sbcl @1.2.11_0+a_sbcl sbcl @1.4.9_0+a_sbcl (active) $ port info sbcl sbcl @1.5.9 (lang) Variants: a_sbcl, [+]fancy, html, pdf, threads ... ... $ sudo port install sbcl -fancy +a_sbcl $ port installed sbcl The following ports are currently installed: sbcl @1.1.6_1 sbcl @1.2.11_0+a_sbcl sbcl @1.4.9_0+a_sbcl sbcl @1.5.9_0+a_sbcl (active) $ port info sbcl sbcl @2.0.9 (lang) Variants: a_sbcl, [+]fancy, html, pdf, threads ... ... $ sudo port install sbcl -fancy +a_sbcl $ port installed sbcl The following ports are currently installed: sbcl @1.1.6_1 sbcl @1.2.11_0+a_sbcl sbcl @1.4.9_0+a_sbcl sbcl @1.5.9_0+a_sbcl sbcl @2.0.9_0+a_sbcl (active)
comment:20 follow-up: 23 Changed 2 years ago by barracuda156
Replying to tomio-arisaka:
Thank you very much! I will look through these.
Do I get it right that I have built without threads and fancy variants? (For me these fail, and at least one reason I point above – a chunk of code invalid for Darwin ABI.)
comment:21 follow-up: 22 Changed 2 years ago by barracuda156
Replying to barracuda156:
Replying to easye:
- There is a section in
src/runtime/ppc-assem.Swhich has never been fixed for Darwin ABI:/* Call out to obtain our TLS block. */ load(reg_NL0,CSYMBOL(specials)) /* This won't work on darwin: wrong fixup style. And is it * supposed to be lis/ori or lis/addi? Or does it differ * between darwin and everything else again? */ lis reg_CFUNC,CSYMBOL(pthread_getspecific)@h ori reg_CFUNC,reg_CFUNC,CSYMBOL(pthread_getspecific)@l mtctr reg_CFUNC bctrl mr reg_THREAD, reg_NL0
There is some info on pthread_getspecific here: https://lifecs.likai.org/2010/05/mac-os-x-thread-local-storage.html
comment:22 Changed 2 years ago by tomio-arisaka (Tomio Arisaka)
Replying to barracuda156:
There is some info on
pthread_getspecifichere: https://lifecs.likai.org/2010/05/mac-os-x-thread-local-storage.html
On Tiger, gcc4 included in Xcode-2.5 cannot compile 'tls.c'. But gcc7 can compile 'tls.c' without any errors:
$ cat ./tls.c
static int __thread x;
int main() { return 0; }
$
$ gcc-mp-7 --version
gcc-mp-7 (MacPorts gcc7 7.5.0_3) 7.5.0
Copyright (C) 2017 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
$
$ gcc-mp-7 ./tls.c
$
$ otool -L ./a.out
./a.out:
/usr/lib/libgcc_s.1.dylib (compatibility version 1.0.0, current version 1.0.0)
/opt/local/lib/libgcc/libgcc_s.1.dylib (compatibility version 1.0.0, current version 1.0.0)
/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 88.1.12)
$
comment:23 follow-up: 26 Changed 2 years ago by tomio-arisaka (Tomio Arisaka)
Replying to barracuda156:
Replying to tomio-arisaka:
Thank you very much! I will look through these.
Do I get it right that I have built without threads and fancy variants? (For me these fail, and at least one reason I point above – a chunk of code invalid for Darwin ABI.)
The implementation of SBCL's threads has not been applied to PPC-Darwin.
So I gave up using SBCL's threads on PPC-Darwin.
But I'm glad that SBCL-2.0.9 can compile Maxima on PPC-Tiger.
By the way, I'm trying to use another Lisp which has the functions of threads. (e.g. ECL-21.2.1 and CCL-1.2/MCL-6.0)
comment:24 Changed 2 years ago by barracuda156
Replying to easye:
I will have to take a break from this (and everything else) until the end of the month, as I will be away from PPC hardware. (Just letting you know that I did not give up.)
comment:25 Changed 23 months ago by barracuda156
Replying to easye:
Support for Darwin/PPC was removed with sbcl-2.2.3 <http://sbcl.org/news.html>.
Gonna be kinda tricky to do this for a single Portfile, so I wonder if anyone can chime in on what should be done: a "new" port with the architecture in the name ('lang/sbcl-ppc')?
We got two alternative pathways now, though both lead to 2.0.9 maximum, at the moment. Which, do you think, is more Macports-compatible?
And if you were able to get any hints from the upstream in the meanwhile, please update us. I can bring back ppc support into the latest sbcl, but we need to fix 2.2.2 first.
comment:26 Changed 23 months ago by barracuda156
Replying to tomio-arisaka:
The implementation of SBCL's threads has not been applied to PPC-Darwin. So I gave up using SBCL's threads on PPC-Darwin.
Thank you for clarifying this!
But I'm glad that SBCL-2.0.9 can compile Maxima on PPC-Tiger.
Good to know that. Hopefully, we at least can bring 2.0.9 for PPC into Macports soon.
By the way, I'm trying to use another Lisp which has the functions of threads. (e.g. ECL-21.2.1 and CCL-1.2/MCL-6.0)
I got clisp built on 10.6 PPC but did not try using it yet.
comment:27 Changed 22 months ago by barracuda156
Replying to easye:
Replying to easye:
Support for Darwin/PPC was removed with sbcl-2.2.3 <http://sbcl.org/news.html>.
A statement from the SBCL maintainer via IRC:
<Krystof> easye: I think it's worth saying that if there are users willing to step up and maintain our ppc/darwin support, we'd reinstate it on that basis [13:20] <Krystof> the problem we were having is: no hardware, no access to hardware, no demand from users, no ability to check whether changes for other platforms cause breakage [13:21]
Resurrecting the thread, where do we stand at the moment? I can restore PPC into the current code, for the most part, however there are two problems: a) threads supports apparently was always broken and b) versions of sbcl past 2.0.9 fail to build. Some advice on the errors I faced (described above) would be very much helpful, since otherwise it is not clear what to try fixing :)
It’s understandable that the upstream cannot test PPC builds, but if it is possible to consult with someone, that gonna be great. For example, what has changed in 2.0.10 substantially enough as to break PPC build?
comment:28 Changed 22 months ago by kakuhen
| Cc: | kakuhen added |
|---|
comment:29 follow-up: 30 Changed 22 months ago by kakuhen
Although performed outside of a MacPorts build environment, I was able to successfully compile SBCL 2.2.2 using SBCL 1.0.47 on Mac OS X 10.5.8. I will attach an archive to this ticket. Attached to this archive is the full compilation together with the test suite and "port installed" output.
I do not have the time right now to attempt patching lang/sbcl/Portfile, but this demonstrates a proof of concept.
Changed 22 months ago by kakuhen
| Attachment: | SBCL LOG.tar.gz added |
|---|
Compile logs of SBCL 2.2.2 on PowerPC Leopard, with TEST results and "port installed" output.
comment:30 Changed 22 months ago by barracuda156
Replying to kakuhen:
Although performed outside of a MacPorts build environment, I was able to successfully compile SBCL 2.2.2 using SBCL 1.0.47 on Mac OS X 10.5.8. I will attach an archive to this ticket. Attached to this archive is the full compilation together with the test suite and "port installed" output.
I do not have the time right now to attempt patching lang/sbcl/Portfile, but this demonstrates a proof of concept.
Thank you very much! I will look into logs and try to figure out how we can patch the Portfile. If that works, then we can restore PPC support into 2.2.6, perhaps.
comment:31 Changed 22 months ago by acjones8 (Alex Jones)
| Cc: | acjones8 added |
|---|
comment:32 Changed 19 months ago by amock (Alan Mock)
| Cc: | amock added |
|---|
comment:33 follow-up: 36 Changed 15 months ago by barracuda156
I have built sbcl 2.3.0 up to this point, and there it freezes:
; SB-Loader: (166+3670) methods/other
SB-XC:*FEATURES* =
(:PPC :GENCGC :ALIEN-CALLBACKS :ANSI-CL :ANSI-COMPLIANT-LOAD-TRUENAME
:BIG-ENDIAN :BSD :COMMON-LISP :COMPARE-AND-SWAP-VOPS :DARWIN
:IEEE-FLOATING-POINT :MACH-O :OS-PROVIDES-BLKSIZE-T :OS-PROVIDES-DLADDR
:OS-PROVIDES-DLOPEN :OS-PROVIDES-SUSECONDS-T :PACKAGE-LOCAL-NICKNAMES :SB-DOC
:SB-EVAL :SB-LDB :SB-PACKAGE-LOCKS :SB-SOURCE-LOCATIONS :SB-UNICODE :SBCL
:UNIX)
[building initial core file in "output/cold-sbcl.core":
writing 4096 bytes [1 page] from #<SB-FASL::GSPACE @#x4100000 :STATIC>
writing 28422144 bytes [6939 pages] from #<SB-FASL::GSPACE @#x10000000 :DYNAMIC>
writing 0 bytes [0 pages] from #<SB-FASL::GSPACE @#x4000000 :READ-ONLY>
movable dynamic space: 534 + 3383 + 3022 cons/code/mixed pages
/INITIAL-FUN=#X119FC1B5
done]
//testing for consistency of first and second GENESIS passes
//header files match between first and second GENESIS -- good
real 9m16.123s
user 7m7.247s
sys 1m22.682s
//entering make-target-2.sh
//doing warm init - compilation phase
This is SBCL 2.3.0, an implementation of ANSI Common Lisp.
More information about SBCL is available at <http://www.sbcl.org/>.
SBCL is free software, provided as is, with absolutely no warranty.
It is mostly in the public domain; some portions are provided under
BSD-style licenses. See the CREDITS and COPYING files in the
distribution for more information.
Initial page table:
Immobile Object Counts
Gen layout fdefn symbol code Boxed Cons Raw Code SmMix Mixed LgRaw LgCode LgMix Waste% Alloc Trig Dirty GCs Mem-age
6 0 0 0 0 0 534 0 3383 0 3022 0 0 0 0.5 28271184 2000000 3383 0 0.0000
Tot 0 0 0 0 0 534 0 3383 0 3022 0 0 0 0.5 28271184 [5.3% of 536870912 max]
Why could that be?
comment:34 Changed 15 months ago by barracuda156
Above was with 1.5.9; I tried now to build with ancient 1.0.47, and it fails with a variation of a known error:
; file: /opt/local/var/macports/build/_opt_PPCSnowLeopardPorts_lang_sbcl/sbcl/work/sbcl-2.3.0/src/compiler/generic/genesis.lisp ; in: DEFUN (SETF BVREF-WORD) ; (SETF (SB-FASL::ACCESS SB-FASL::BYTES SB-INT:INDEX T) SB-FASL::NEW-VAL) ; ; caught ERROR: ; during macroexpansion of (SETF # NEW-VAL). Use *BREAK-ON-SIGNALS* to intercept: ; ; The assertion (AND SB-COLD::RESULT (NOT (CDR SB-COLD::RESULT))) failed. ; (DEFUN (SETF SB-FASL::BVREF-WORD) ; (SB-FASL::NEW-VAL SB-FASL::BYTES SB-INT:INDEX) ; (DECLARE (TYPE SB-XC:FIXNUM SB-INT:INDEX)) ; (SB-INT:AVER (NOT (LOGTEST SB-INT:INDEX (ASH SB-VM:LOWTAG-MASK -1)))) ; (SETF (SB-FASL::ACCESS SB-FASL::BYTES SB-INT:INDEX T) SB-FASL::NEW-VAL)) ; --> PROGN EVAL-WHEN ; ==> ; (HOST-SB-IMPL::%DEFUN '(SETF SB-FASL::BVREF-WORD) ; (HOST-SB-INT:NAMED-LAMBDA (SETF SB-FASL::BVREF-WORD) ; (SB-FASL::NEW-VAL SB-FASL::BYTES SB-INT:INDEX) ; (DECLARE (TYPE SB-XC:FIXNUM SB-INT:INDEX)) ; (BLOCK SB-FASL::BVREF-WORD ; (SB-INT:AVER (NOT #)) ; (SETF # SB-FASL::NEW-VAL))) ; NIL 'NIL (HOST-SB-C:SOURCE-LOCATION)) ; ; caught STYLE-WARNING: ; The variable NEW-VAL is defined but never used. ; ; caught STYLE-WARNING: ; The variable BYTES is defined but never used. ; in: DEFUN BVREF-WORD ; (SB-FASL::ACCESS SB-FASL::BYTES SB-INT:INDEX T) ; ; caught ERROR: ; during macroexpansion of (ACCESS BYTES INDEX ...). Use *BREAK-ON-SIGNALS* to ; intercept: ; ; The assertion (AND SB-COLD::RESULT (NOT (CDR SB-COLD::RESULT))) failed. ; (DEFUN SB-FASL::BVREF-WORD (SB-FASL::BYTES SB-INT:INDEX) ; (DECLARE (TYPE SB-XC:FIXNUM SB-INT:INDEX)) ; (SB-INT:AVER (NOT (LOGTEST SB-INT:INDEX (ASH SB-VM:LOWTAG-MASK -1)))) ; (SB-FASL::ACCESS SB-FASL::BYTES SB-INT:INDEX T)) ; --> PROGN EVAL-WHEN ; ==> ; (HOST-SB-IMPL::%DEFUN 'SB-FASL::BVREF-WORD ; (HOST-SB-INT:NAMED-LAMBDA SB-FASL::BVREF-WORD ; (SB-FASL::BYTES SB-INT:INDEX) ; (DECLARE (TYPE SB-XC:FIXNUM SB-INT:INDEX)) ; (BLOCK SB-FASL::BVREF-WORD ; (SB-INT:AVER (NOT #)) ; (SB-FASL::ACCESS SB-FASL::BYTES SB-INT:INDEX T))) ; NIL 'NIL (HOST-SB-C:SOURCE-LOCATION)) ; ; caught STYLE-WARNING: ; The variable BYTES is defined but never used. ; /opt/local/var/macports/build/_opt_PPCSnowLeopardPorts_lang_sbcl/sbcl/work/sbcl-2.3.0/obj/from-host/src/compiler/generic/genesis.fasl-tmp written ; compilation finished in 0:00:11.403 debugger invoked on a SIMPLE-ERROR: FAILURE-P was set when creating "obj/from-host/src/compiler/generic/genesis.fasl". Type HELP for debugger help, or (SB-EXT:QUIT) to exit from SBCL. restarts (invokable by number or by possibly-abbreviated name): 0: [RECOMPILE ] Recompile file "src/compiler/generic/genesis.lisp" 1: [CONTINUE ] Continue, using possibly bogus file "obj/from-host/src/compiler/generic/genesis.fasl" 2: Recompile 3: [ABORT-BUILD] Abort building SBCL. 4: [ABORT ] Exit debugger, returning to top level. (COMPILE-STEM "src/compiler/generic/genesis" NIL :HOST-COMPILE) 0] 1 Command failed: SBCL_MACOSX_VERSION_MIN=10.6 CC=/opt/local/bin/gcc-mp-12 CXX=/opt/local/bin/g++-mp-12 CPP=/opt/local/bin/cpp-mp-12 sh ./make.sh --prefix=/opt/local --xc-host="/opt/local/var/macports/build/_opt_PPCSnowLeopardPorts_lang_sbcl/sbcl/work/sbcl-1.0.47-powerpc-darwin/src/runtime/sbcl --core /opt/local/var/macports/build/_opt_PPCSnowLeopardPorts_lang_sbcl/sbcl/work/sbcl-1.0.47-powerpc-darwin/output/sbcl.core --sysinit /dev/null --userinit /dev/null" Exit code: 128 Error: Failed to build sbcl: command execution failed Error: See /opt/local/var/macports/logs/_opt_PPCSnowLeopardPorts_lang_sbcl/sbcl/main.log for details. Error: Follow https://guide.macports.org/#project.tickets if you believe there is a bug.
comment:35 Changed 14 months ago by kencu (Ken)
| Keywords: | powerpc added |
|---|---|
| Summary: | sbcl: fatal error encountered in SBCL pid 16153: dynamic space too small for core → sbcl fails to build on PPC: fatal error encountered in SBCL pid 16153: dynamic space too small for core |
comment:36 Changed 12 months ago by barracuda156
Replying to barracuda156:
I have built
sbcl2.3.0 up to this point, and there it freezes:; SB-Loader: (166+3670) methods/other SB-XC:*FEATURES* = (:PPC :GENCGC :ALIEN-CALLBACKS :ANSI-CL :ANSI-COMPLIANT-LOAD-TRUENAME :BIG-ENDIAN :BSD :COMMON-LISP :COMPARE-AND-SWAP-VOPS :DARWIN :IEEE-FLOATING-POINT :MACH-O :OS-PROVIDES-BLKSIZE-T :OS-PROVIDES-DLADDR :OS-PROVIDES-DLOPEN :OS-PROVIDES-SUSECONDS-T :PACKAGE-LOCAL-NICKNAMES :SB-DOC :SB-EVAL :SB-LDB :SB-PACKAGE-LOCKS :SB-SOURCE-LOCATIONS :SB-UNICODE :SBCL :UNIX) [building initial core file in "output/cold-sbcl.core": writing 4096 bytes [1 page] from #<SB-FASL::GSPACE @#x4100000 :STATIC> writing 28422144 bytes [6939 pages] from #<SB-FASL::GSPACE @#x10000000 :DYNAMIC> writing 0 bytes [0 pages] from #<SB-FASL::GSPACE @#x4000000 :READ-ONLY> movable dynamic space: 534 + 3383 + 3022 cons/code/mixed pages /INITIAL-FUN=#X119FC1B5 done] //testing for consistency of first and second GENESIS passes //header files match between first and second GENESIS -- good real 9m16.123s user 7m7.247s sys 1m22.682s //entering make-target-2.sh //doing warm init - compilation phase This is SBCL 2.3.0, an implementation of ANSI Common Lisp. More information about SBCL is available at <http://www.sbcl.org/>. SBCL is free software, provided as is, with absolutely no warranty. It is mostly in the public domain; some portions are provided under BSD-style licenses. See the CREDITS and COPYING files in the distribution for more information. Initial page table: Immobile Object Counts Gen layout fdefn symbol code Boxed Cons Raw Code SmMix Mixed LgRaw LgCode LgMix Waste% Alloc Trig Dirty GCs Mem-age 6 0 0 0 0 0 534 0 3383 0 3022 0 0 0 0.5 28271184 2000000 3383 0 0.0000 Tot 0 0 0 0 0 534 0 3383 0 3022 0 0 0 0.5 28271184 [5.3% of 536870912 max]Why could that be?
36-55% sudo gdb-apple attach 82530 Password: GNU gdb 6.3.50.20050815-cvs (Wed Apr 13 04:55:50 UTC 2022) Copyright 2004 Free Software Foundation, Inc. GDB is free software, covered by the GNU General Public License, and you are welcome to change it and/or distribute copies of it under certain conditions. Type "show copying" to see the conditions. There is absolutely no warranty for GDB. Type "show warranty" for details. This GDB was configured as "--host=powerpc-apple-darwin10.0.0d2 --target="...File 'attach' does not exist or is not a regular file. /Users/svacchanda/82530: No such file or directory Attaching to process 82530. Reading symbols for shared libraries . done Reading symbols for shared libraries ... done 0xbffff410 in ?? () (gdb) next Cannot find bounds of current function (gdb) where #0 0xbffff410 in ?? () #1 0x0002cc78 in _call_into_c () at ppc-assem.S:531 #2 0x000012e4 in ?? () #3 0x8fe3342c in __dyld__ZN4dyld8gLogAPIsE () #4 0x8fe0c558 in __dyld_dlsym () #5 0x0001a484 in create_main_lisp_thread (function=1110249391) at thread.c:375 #6 0x0001708c in initialize_lisp (argc=<value temporarily unavailable, due to optimizations>, argv=0xc24bc195, envp=<value temporarily unavailable, due to optimizations>) at runtime.c:813 #7 0x00034f7c in main (argc=<value temporarily unavailable, due to optimizations>, argv=<value temporarily unavailable, due to optimizations>, envp=<value temporarily unavailable, due to optimizations>) at main.c:6 (gdb) disassemble $pc-0x10 $pc+0x10 Dump of assembler code from 0xbffff400 to 0xbffff420: 0xbffff400: .long 0x35dc4 0xbffff404: .long 0x35dc0 0xbffff408: stmw r31,-2220(r31) 0xbffff40c: lbzu r31,14012(r3) 0xbffff410: .long 0x11a54830 0xbffff414: .long 0x35dc0 0xbffff418: stmw r31,-2220(r31) 0xbffff41c: lbzu r31,14012(r3) End of assembler dump. (gdb) info registers r0 0x0 0 r1 0xbffff390 3221222288 r2 0x364fc 222460 r3 0x2 2 r4 0x11a54830 296044592 r5 0xd 13 r6 0x0 0 r7 0x4000030 67108912 r8 0x11a54830 296044592 r9 0x11a54830 296044592 r10 0x34 52 r11 0x0 0 r12 0xbffff410 3221222416 r13 0x330 816 r14 0x1200008 18874376 r15 0x1000108 16777480 r16 0x1000128 16777512 r17 0x1400000 20971520 r18 0x4100033 68157491 r19 0x10816007 276914183 r20 0x0 0 r21 0x34 52 r22 0x10000e8 16777448 r23 0x108161ff 276914687 r24 0x8 8 r25 0x11a5482f 296044591 r26 0x0 0 r27 0x34 52 r28 0xbffff3f0 3221222384 r29 0x11a5482f 296044591 r30 0x4100033 68157491 r31 0x10816330 276914992 pc 0xbffff410 3221222416 ps 0x100000000008d030 1152921504607424560 cr 0x24002200 603988480 lr 0x2cc78 183416 ctr 0xbffff410 3221222416 xer 0x0 0 mq 0x0 0 fpscr 0x82002000 2181046272 vscr 0x10000 65536 vrsave 0x0 0
comment:37 Changed 11 months ago by catap (Kirill A. Korinsky)
| Resolution: | → fixed |
|---|---|
| Status: | assigned → closed |
For the record, current test status on PPC.
sbcl1.5.9 (tested on 10.6 PPC):sbcl1.2.8 (tested on 10.5.8 and 10.6 PPC):