Singular fails to build in Sage 5.9

[Pong]

Looking for help in compiling SAGE 5.9

I encountered an error in buliding Singular-3-1-5.p7

I suspect the following may be the relevant part of the log file

In file included from ../kernel/si_gmp.h:4:0,
                 from ../kernel/structs.h:15,
                 from weight0.c:13:
../factory/cf_gmp.h:2501:28: fatal error: bits/c++config.h: No such file or directory
 #include <bits/c++config.h>
                            ^
compilation terminated.
make[4]: *** [weight0.o] Error 1

However, I also attached the full file for the full details.
Thank you for the help in advance.

[Francois Bissey]

A fuller section of the log:
gcc -O2 -g -fPIC -pipe -I. -I.. -I/home/pong/sage-5.9/local
-I/home/pong/sage-5.9/local/include -I/home/pong/sage-5.9/local/include
-I/home/pong/sage-5.9/local/include -I/usr/local/include -DNDEBUG
-DOM_NDEBUG -Dx86_64_Linux -DHAVE_CONFIG_H -c weight0.c

In file included from ../kernel/si_gmp.h:4:0,
from ../kernel/structs.h:15,
from weight0.c:13:
../factory/cf_gmp.h:2501:28: fatal error: bits/c++config.h: No such file
or directory
#include <bits/c++config.h>

so bits/c++config.h is a c++ header provided by the compiler in your
case a pre-release of gcc-4.8.0. But the killer thing I think is that
weight0.c is a justly compiled with the c compiler but if you want
to use this header, g++ should be used.
One man's opinion at any rate.

Francois

[Wai Yan Pong]

Hum I see what you are saying. Well my case is a bit tricky but I think I'm not alone. I am using Archlinux, so there is not g++ separately. And gcc decides whether the program as a C or C++ program in the following way.

gcc file.C # uppercase .C (or .cpp) will process as C++ 
gcc file.c # lowercase .c will process as C

see http://unix.stackexchange.com/questions/34654/c-compile-in-arch-linux-with-no-g-package-in-repositories for details.

So the question is what can be changed so that the gcc in my machine will interpret compile weight0.c as a C++ program... 

I wonder why I didn't encounter such a program earlier... perhaps it's the pre-release reversion that causes that problem. 

Also can I use force it to use the gcc (I suppose gcc-4.7.2.p1.spkg is the one) that comes with sage to compile? 

Francois

--
You received this message because you are subscribed to a topic in the Google Groups "sage-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sage-devel/hqH8t69ixxA/unsubscribe?hl=en.
To unsubscribe from this group and all its topics, send an email to sage-devel+...@googlegroups.com.
To post to this group, send email to sage-...@googlegroups.com.
Visit this group at http://groups.google.com/group/sage-devel?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.

[Francois Bissey]

On 05/05/13 21:12, Wai Yan Pong wrote:
> Hum I see what you are saying. Well my case is a bit tricky but I think
> I'm not alone. I am using Archlinux, so there is not g++ separately. And
> gcc decides whether the program as a C or C++ program in the following way.
>
> |gcc file.C # uppercase .C (or .cpp) will process as C++
> gcc file.c # lowercase .c will process as C|
>
> see
> http://unix.stackexchange.com/questions/34654/c-compile-in-arch-linux-with-no-g-package-in-repositories
> for details.
>
> So the question is what can be changed so that the gcc in my machine
> will interpret compile weight0.c as a C++ program...
> I wonder why I didn't encounter such a program earlier... perhaps it's
> the pre-release reversion that causes that problem.
> Also can I use force it to use the gcc (I suppose gcc-4.7.2.p1.spkg is
> the one) that comes with sage to compile?
>

Yes it looks tricky. It could very well be a regression as it compile
with the C compiler and not the C++ compiler on my box with gcc 4.6.3
(gentoo). That being said gcc is very permissive and one day someone may
decide to cut that one off. Would you believe that 2 days ago I
accidentally configured octave with CXX=gcc-4.7.1 and I almost
completely compiled it before being shot down by such a problem [octave
is mostly c++].

Anyway you can certainly force sage's gcc by setting
SAGE_INSTALL_GCC=yes before building sage.

Francois

[Pong]

Hi Francois, it's a relief to learn that you run into similar problems.

Well I followed your suggestion and set SAGE_INSTALL_GCC to yes and then rebuilt. Unfortunately, ran into the same error.
Looks like it still try to use my system's GCC install of the one in SAGE-5.9

Running out of idea, I wipped out the directory and started again. This time I tried to build gcc-4.7.2.p1.spkg first (perhaps that the wrong step).
I issued

./sage -f gcc-4.7.2.p1.spkg

but that failed to build either! I attached the log file for people to investigate.
I tried setting the variable SAGE_INSTALL_GCC either way too.

[leif]

The problem is not that a C compiler is used to compile weight0.c, but
that cf_gmp.h includes <bits/c++config.h> in the first place.

(Although bits/c++config.h is valid C code as well, it's not in the
[C++] include path if you compile a file as C code which includes it.)

But it seems ArchLinux's broken gcc/g++ behaviour triggers the problem.


Pong, could you upload/attach your cf_gmp.h? (Should be located in
$SAGE_ROOT/spkg/build/singular-3-1-5.p7/src/factory/, but only after a
failed build or if you use 'sage -f -s ...', or alternatively set
SAGE_KEEP_BUILT_SPKGS=yes.)


I'll take a look at the full Singular build log you attached earlier.


-leif

--
() The ASCII Ribbon Campaign
/\ Help Cure HTML E-Mail

[Pong]

Hi leif, thank you for looking into this issue as well.
Attached is my cf_gmp.h after a failed build attempt with sage -f singular-3-1-5.p7

Thanks

[leif]

leif wrote:
> The problem is not that a C compiler is used to compile weight0.c, but
> that cf_gmp.h includes <bits/c++config.h> in the first place.
>
> (Although bits/c++config.h is valid C code as well, it's not in the
> [C++] include path if you compile a file as C code which includes it.)
>
> But it seems ArchLinux's broken gcc/g++ behaviour triggers the problem.

Though it might still be a genuine Singular bug.

> Pong, could you upload/attach your cf_gmp.h? (Should be located in
> $SAGE_ROOT/spkg/build/singular-3-1-5.p7/src/factory/, but only after a
> failed build or if you use 'sage -f -s ...', or alternatively set
> SAGE_KEEP_BUILT_SPKGS=yes.)
>
>
> I'll take a look at the full Singular build log you attached earlier.

The relevant lines seem to be these:

echo "#!/bin/sh" >gen_cf_gmp.sh
echo "GMP_H_T=\"`g++ -Wall -fno-implicit-templates -I. -I.. -I.
-I/home/pong/sage-5.9/local -I/home/pong/sage-5.9/local/include
-DHAVE_CONFIG_H -I/home/pong/sage-5.9/local/include
-I/home/pong/sage-5.9/local/include -O2 -g -fPIC -M
gen_cf_gmp.cc|grep gmp.h` \"" >>gen_cf_gmp.sh
cat gen_cf_gmp.template >>gen_cf_gmp.sh
/bin/sh gen_cf_gmp.sh
generating cf_gmp.h from /home/pong/sage-5.9/local/include/gmp.h
/usr/include/c++/4.8.0/iosfwd

(Note the iosfwd.)

[leif]

Pong wrote:
> Hi leif, thank you for looking into this issue as well.
> Attached is my cf_gmp.h after a failed build attempt with sage -f
> singular-3-1-5.p7

Well, that version doesn't include bits/c++config.h. (It doesn't have
as many lines as the previous one where the error occurred, and at first
glance looks pretty sane.)

Do you get a different error now?


-leif

[Pong]

Nope. I got the exact seem error. See the lastest log attached.

One more thing that puzzled me is that the arch users just got an update (including my machine) on gcc. So it's not the pre-release any more.
~ [i]> pacman -Qi gcc
Name           : gcc
Version        : 4.8.0-4
Description    : The GNU Compiler Collection - C and C++ frontends
Architecture   : x86_64

But somehow in the logs it still set I'm using the prerelease compiler. I'm clueless.

[François Bissey]

On Sun, 05 May 2013 16:07:11 Pong wrote:
> Nope. I got the exact seem error. See the lastest log attached.
>
> One more thing that puzzled me is that the arch users just got an update
> (including my machine) on gcc. So it's not the pre-release any more.
> ~ [i]> pacman -Qi gcc
> Name : gcc
> Version : 4.8.0-4
> Description : The GNU Compiler Collection - C and C++ frontends
> Architecture : x86_64
>

And if you do "gcc --version" what does it says?

Francois

[Pong]

Well... it said "prerelease"
gcc version 4.8.0 20130502 (prerelease) (GCC)

I see now, obviously 4.8.0 is still a prerelease...

Just out of curiousity, I have just downloaded a copy of Singular-3.1.5 from the offical site and see if it compiles on my machine.
Will keep everyone posted.

[Pong]

Report: buliding Singular by itself seems to be completely fine.

./configure
make -j4
results in 
....
/usr/bin/install -c  -s solve_IP /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
/usr/bin/install -c  -s change_cost /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
/usr/bin/install -c  -s toric_ideal /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
/usr/bin/install -c  -s gen_test /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
/usr/bin/install -c  -s LLL /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
make[2]: Leaving directory `/home/pong/Downloads/Singular-3-1-5/IntegerProgramming'
echo "* do not forget to install Singular-3-1-5-share.tar.gz"
* do not forget to install Singular-3-1-5-share.tar.gz
make[1]: Leaving directory `/home/pong/Downloads/Singular-3-1-5'

and it gives me back the prompt. I'm not sure if this "test" is helpful in bugging what's going on. But perhaps it is not a Singular bug.

[leif]

Pong wrote:
> Report: buliding Singular by itself seems to be completely fine.
>
> ./configure
> make -j4
> results in
> ....

Well, you'd have to (at least) configure with '--with-apint=gmp' as well
(cf. Sage's spkg-install script).


-leif

> /usr/bin/install -c -s solve_IP
> /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
> /usr/bin/install -c -s change_cost
> /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
> /usr/bin/install -c -s toric_ideal
> /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
> /usr/bin/install -c -s gen_test
> /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
> /usr/bin/install -c -s LLL /home/pong/Downloads/Singular-3-1-5/x86_64-Linux
> make[2]: Leaving directory
> `/home/pong/Downloads/Singular-3-1-5/IntegerProgramming'
> echo "* do not forget to install Singular-3-1-5-share.tar.gz"
> * do not forget to install Singular-3-1-5-share.tar.gz
> make[1]: Leaving directory `/home/pong/Downloads/Singular-3-1-5'
>
> and it gives me back the prompt. I'm not sure if this "test" is helpful
> in bugging what's going on. But perhaps it is not a Singular bug.
>
> On Sunday, May 5, 2013 4:31:10 PM UTC-7, Pong wrote:
>
> Well... it said "prerelease"
> gcc version 4.8.0 20130502 (prerelease) (GCC)
>
> I see now, obviously 4.8.0 is still a prerelease...
>
> Just out of curiousity, I have just downloaded a copy of
> Singular-3.1.5 from the offical site and see if it compiles on my
> machine.
> Will keep everyone posted.

[Pong]

Oh... I see. I am ignorant on what various flags mean.

Let me summarize my problems and see if I get them right:

1) my gcc compiler confuses c and c++ programs.
2) I want to solve 1) by forcing sage to use the gcc-4.7.2. but the problem is that the copy of gcc that come with sage won't build from my system gcc...
    (see the my 3rd message in this discussion)

Am I right? Any ideas on what else can I try?

[Francois Bissey]

On 06/05/13 17:24, Pong wrote:
> Oh... I see. I am ignorant on what various flags mean.
>
> Let me summarize my problems and see if I get them right:
>
> 1) my gcc compiler confuses c and c++ programs.
> 2) I want to solve 1) by forcing sage to use the gcc-4.7.2. but the
> problem is that the copy of gcc that come with sage won't build from my
> system gcc...
> (see the my 3rd message in this discussion)
>
> Am I right? Any ideas on what else can I try?
>

Have you tried a build from scratch with the settings to buid
sage's own gcc?

Francois

[Pong]

Does it simply mean set the variable SAGE_INSTALL_GCC to yes and then issue "make" in the directory newly created by tar xvf... ?
 

[Francois Bissey]

yes.

Francois

--
You received this message because you are subscribed to the Google Groups "sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.

[Pong]

Alright, I did a fresh compile with SAGE_INSTALL_GCC set to yes. This time the compilation failed at gcc.
It only installed a few things before it hit gcc (e.g. Singular was not complied)
 
And here are the last few words it said before the crash:

../../src/gcc/doc/cppopts.texi:806: @itemx must follow @item
make[6]: *** [doc/cpp.info] Error 1
make[6]: Leaving directory `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build/gcc'
make[5]: *** [all-stage1-gcc] Error 2
make[5]: Leaving directory `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build'
make[4]: *** [stage1-bubble] Error 2
make[4]: Leaving directory `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build'
make[3]: *** [all] Error 2
make[3]: Leaving directory `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build'

Please see the attached log for details.

On Monday, May 6, 2013 11:21:30 AM UTC-7, François wrote:

yes.

Francois

On 7/05/2013, at 3:41, "Pong" <wypo...@gmail.com> wrote:

On Monday, May 6, 2013 2:19:47 AM UTC-7, François wrote:

On 06/05/13 17:24, Pong wrote:
> Oh... I see. I am ignorant on what various flags mean.
>
> Let me summarize my problems and see if I get them right:
>
> 1) my gcc compiler confuses c and c++ programs.
> 2) I want to solve 1) by forcing sage to use the gcc-4.7.2. but the
> problem is that the copy of gcc that come with sage won't build from my
> system gcc...
>     (see the my 3rd message in this discussion)
>
> Am I right? Any ideas on what else can I try?
>

Have you tried a build from scratch with the settings to buid
sage's own gcc?

Francois

Does it simply mean set the variable SAGE_INSTALL_GCC to yes and then issue "make" in the directory newly created by tar xvf... ?
 

--
You received this message because you are subscribed to the Google Groups "sage-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.

To post to this group, send email to sage...@googlegroups.com.

[François Bissey]

On Mon, 06 May 2013 20:39:26 Pong wrote:
> Alright, I did a fresh compile with SAGE_INSTALL_GCC set to yes. This time
> the compilation failed at gcc.
> It only installed a few things before it hit gcc (e.g. Singular was not
> complied)
>
> And here are the last few words it said before the crash:
>
> ../../src/gcc/doc/cppopts.texi:806: @itemx must follow @item
> make[6]: *** [doc/cpp.info] Error 1
> make[6]: Leaving directory
> `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build/gcc'
> make[5]: *** [all-stage1-gcc] Error 2
> make[5]: Leaving directory
> `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build'
> make[4]: *** [stage1-bubble] Error 2
> make[4]: Leaving directory
> `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build'
> make[3]: *** [all] Error 2
> make[3]: Leaving directory
> `/home/pong/sage-5.9/spkg/build/gcc-4.7.2.p1/gcc-build'
>
> Please see the attached log for details.
>

Looks like a stupid texinfo error. There has been a lot of these around
lately.

Francois

[Wai Yan Pong]

Hum so does someone know of a fix?

--
You received this message because you are subscribed to a topic in the Google Groups "sage-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sage-devel/hqH8t69ixxA/unsubscribe?hl=en.
To unsubscribe from this group and all its topics, send an email to sage-devel+...@googlegroups.com.
To post to this group, send email to sage-...@googlegroups.com.

[leif]

Pong wrote:
> Oh... I see. I am ignorant on what various flags mean.

Did you try to build the stand-alone/vanilla Singular configured with
'--with-apint=gmp'?

> Let me summarize my problems and see if I get them right:
>
> 1) my gcc compiler confuses c and c++ programs.

I don't think so, at least not immediately. The file whose compilation
fails (using 'gcc') is supposed to be C code (and hence to be compiled
as such), but -- for whatever reason -- the Singular scripts that create
a header file (cf_gmp.h) included by that also (erroneously) include a
file (bits/c++config.h) that is simply not in the (default) search path
for header files when compiling C code.

[I don't know yet whether ArchLinux, Singular, or maybe even Sage is to
blame.]

> 2) I want to solve 1) by forcing sage to use the gcc-4.7.2. but the
> problem is that the copy of gcc that come with sage won't build from my
> system gcc...
> (see the my 3rd message in this discussion)
>
> Am I right? Any ideas on what else can I try?

Unfortunately, Singular's spkg-install is too "complex" to just manually
fix (the generated) src/factory/cf_gmp.h in a failed build and rerun
./spkg-install (from within a Sage subshell of course)... (You could
try rerunning the appropriate 'make' command afterwards though, and see
whether the remaining pieces of that target build. To resume the build
of all of Singular, you'd have to temporarily change spkg-install, or
manually issue all remaining commands, afterwards -- upon success --
manually touching spkg/installed/singular-<version>.)

But you could try to track the issue (i.e., the reason for the invalid
inclusion) further down, also by trying to build vanilla Singular 3-1-5
as mentioned above.


-leif

[Pong]

Hi Leif

    I haven't tried any of those. In fact, I just reinstall SAGE 5.8.1 which is available on the Arch community. Does anyone here know who is maintaining that package? It would be great if we get an update from there soon as well.

    I will give your suggestion a try and hopefully I learn more during the process but that has to wait till the final exams here over (in a week).

[Pong]

Hi arojas,

    Which PKGBUILD did you use? Where I search on the arch repo, I only see SAGE-5.8 (in community) and SAGE-5.10-devel in AUR.
Where I downloaded the source (SAGE-5.9) and fellowed your suggestion by setting CPP to /usr/bin/cpp, I got the exact same error: "bits/c++config.h: No such file or directory"

On Thursday, May 9, 2013 11:48:01 AM UTC-7, arojas wrote:

El jueves, 9 de mayo de 2013 20:21:43 UTC+2, Pong  escribió:
> Hi Leif
>
>
>     I haven't tried any of those. In fact, I just reinstall SAGE 5.8.1 which is available on the Arch community. Does anyone here know who is maintaining that package? It would be great if we get an update from there soon as well.
>

I compiled sage 5.9 successfully in Arch x86_64. I just had to add 'export CPP=/usr/bin/cpp' to the PKGBUILD before compilation

[arojas]

On Saturday, May 11, 2013 10:17:52 AM UTC+2, Pong wrote:

Hi arojas,

    Which PKGBUILD did you use? Where I search on the arch repo, I only see SAGE-5.8 (in community) and SAGE-5.10-devel in AUR.
Where I downloaded the source (SAGE-5.9) and fellowed your suggestion by setting CPP to /usr/bin/cpp, I got the exact same error: "bits/c++config.h: No such file or directory"

Use the one from [community], just update the version number and the md5sum 

[Pong]

The one in community is 5.8 (not 5.9) and isn't it a binary package?
I'm looking at the one from AUR which is 5.10 beta.

[arojas]

On Saturday, May 11, 2013 10:33:01 AM UTC+2, Pong wrote:

The one in community is 5.8 (not 5.9) and isn't it a binary package?
I'm looking at the one from AUR which is 5.10 beta.

Binary packages also have PKGBUILDs. You can download them, modify them as you need and recompile. Check the wiki https://wiki.archlinux.org/index.php/Arch_Build_System 

[Pong]

Alright, I see. Thanks. I only didn't know that. I thought PKGBUILDs are available only to those in AUR, since they ask you explicitly whether you want to edit them during installation.

We are getting somewhere since I compiles through Singular. However, it then failed at the conway_polynomial...
It complains

OSError: [Errno 13] Permission denied: '/build'

I checked the permission of .../sage-mathematics/src/sage-5.9/spkg/build/ and it seems okay.
attached it the logs and the PKGBUILD (I changed the extension to txt, cos google complains about the type and did not allow the upload).

Thanks in advance

[Pong]

Another wierd thing is that, it produces a pkg/ directory with a strange premission setting.

d---------  2 pong users      4096 May 11 09:05 pkg/

I tried to set it to 755 and it is set to that again when I issue makepkg -i

On Saturday, May 11, 2013 2:16:03 AM UTC-7, arojas wrote:

[arojas]

Just change the DOT_SAGE line to point to some dir where you have write permissions, eg /tmp. Sorry, I forgot to warn you about that, but if you still have the build dir around makepkg won't need to rebuild everything again

[Pong]

Thanks. I do still have the prebuild and it passes through the build.
But now it hangs at

Creating sage-5.9-x86_64-Linux.tar.gz ...
Moving final distribution file to /home/pong/sage-mathematics/src/sage-5.9/dist
==> Entering fakeroot environment...
==> Starting package()...
==> Tidying install...
  -> Purging unwanted files...
  -> Compressing man and info pages...
  -> Stripping unneeded symbols from binaries and libraries...
strip:./opt/sage/local/lib/libsingcf_g.a(readcf.y): Unable to recognise the format of file: File format not recognized

while making the tar.xz file. Well, maybe it will take longer but htop does not show anything involved is running...
If it still the case, I will try the whole thing again one last time from scratch.

[Pong]

Thank you all for helping along the way. I got SAGE-5.9 installed finally.

It was just a long wait at the stripping unneeded symbols from binaries and libraries... 

[leif]

Pong wrote:
> Looking for help in compiling SAGE 5.9
>
> I encountered an error in buliding Singular-3-1-5.p7
>
> I suspect the following may be the relevant part of the log file
>
> In file included from ../kernel/si_gmp.h:4:0,
> from ../kernel/structs.h:15,
> from weight0.c:13:
> ../factory/cf_gmp.h:2501:28: fatal error: bits/c++config.h: No such file
> or directory
> #include <bits/c++config.h>
> ^
> compilation terminated.
> make[4]: *** [weight0.o] Error 1

This is an upstream Singular bug; see #14737 [1].


-leif

[1] http://trac.sagemath.org/sage_trac/ticket/14737