Bug#804297: graphviz: dot on mipsel fails with emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed

[GCS]

On Mon, Apr 11, 2016 at 10:37 AM, Hector Oron  wrote:
> Compiling graphviz with -fno-ipa-sra helps and makes the issue go away.
>
> The other option would be to disable graphs.
>
> While GCC upstream decides, I would prefer to build graphviz with 
> -fno-ipa-sra,
> at least for mips/mipsel architectures. Would that be fine with graphviz
> maintainer? Failing that we should fallback to disable graphs in wayland
> package and many others which might need and have already done so.
 I can compile with -fno-ipa-sra and this is not a problem. Will do it
in the afternoon.

Thanks for the tip,
Laszlo/GCS

Bug#804297: graphviz: dot on mipsel fails with emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed

[Hector Oron]

On Fri, Apr 08, 2016 at 01:40:19PM +0200, Emilio Pozuelo Monfort wrote:
> On Tue, 10 Nov 2015 23:19:57 +0100 Aurelien Jarno  
> wrote:

> > Yes, this is due to gcc-5 and compiling emit.c with -fno-ipa-sra
> > workarounds the problem. This is therefore likely the same issue than
> > bug #804536 / upstream bug PR68273, though I haven't analyzed the
> > generated code in details.
 
> That GCC bug is fixed in Debian, and graphviz has been rebuilt. However dot
> still fails, see:

If you carefully read upstream bug, it has only been solved partially.
The upstream failure was described as a gsoap issue, and upstream fixed that,
however, dot failure is related but not fixed upstream.

We successfully tested two different patches (one for backend code, other for
arch code) for GCC that solves the issue, but stream is undecided in which to
apply.

Compiling graphviz with -fno-ipa-sra helps and makes the issue go away.

The other option would be to disable graphs.

While GCC upstream decides, I would prefer to build graphviz with -fno-ipa-sra,
at least for mips/mipsel architectures. Would that be fine with graphviz
maintainer? Failing that we should fallback to disable graphs in wayland
package and many others which might need and have already done so.

Regards,
-- 
  Hector Oron


signature.asc
Description: PGP signature

Bug#804297: graphviz: dot on mipsel fails with emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed

[Emilio Pozuelo Monfort]

On Tue, 10 Nov 2015 23:19:57 +0100 Aurelien Jarno  wrote:
> On 2015-11-09 11:38, James Cowgill wrote:
> > Control: notfound -1 2.38.0-10
> > Control: found -1 2.38.0-11
> > 
> > Hi,
> > 
> > On Sat, 2015-11-07 at 03:52 +0100, Johannes Schauer wrote:
> > > Package: graphviz
> > > Version: 2.38.0-10
> > > Severity: important
> > > Control: block 804296 by -1
> > > 
> > > Hi,
> > > 
> > > I ran into this problem when building my package botch on mipsel:
> > [...]
> > > The problem can be trigged by the most minimal input:
> > > 
> > >   > $ echo -ne 'digraph G {A -> B}' | dot
> > >   > dot: emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed.
> > >   Aborted
> > 
> > I can reproduce this using graphviz 2.38.0-11, but not 2.38.0-10.
> > 
> > The most obvious difference between them is that -11 is compiled with
> > gcc-5 and -10 is not. That could be related (I havn't done much
> > investigation so far)?
> 
> Yes, this is due to gcc-5 and compiling emit.c with -fno-ipa-sra
> workarounds the problem. This is therefore likely the same issue than
> bug #804536 / upstream bug PR68273, though I haven't analyzed the
> generated code in details.

That GCC bug is fixed in Debian, and graphviz has been rebuilt. However dot
still fails, see:

https://buildd.debian.org/status/fetch.php?pkg=wayland=mips=1.10.0-1=1460114668

/usr/bin/dot -Tpng -oxml/wayland-architecture.png
../../../doc/doxygen/dot/wayland-architecture.gv
Makefile:604: recipe for target 'xml/wayland-architecture.png' failed
make[4]: *** [xml/wayland-architecture.png] Segmentation fault

https://buildd.debian.org/status/fetch.php?pkg=wayland=mipsel=1.10.0-1=1460114941

/usr/bin/dot -Tpng -oxml/wayland-architecture.png
../../../doc/doxygen/dot/wayland-architecture.gv
dot: emit.c:3874: bezier_bb: Assertion `bz.size % 3 == 1' failed.
Makefile:604: recipe for target 'xml/wayland-architecture.png' failed

Cheers,
Emilio

Bug#804297: graphviz: dot on mipsel fails with emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed

[Aurelien Jarno]

On 2015-11-09 11:38, James Cowgill wrote:
> Control: notfound -1 2.38.0-10
> Control: found -1 2.38.0-11
> 
> Hi,
> 
> On Sat, 2015-11-07 at 03:52 +0100, Johannes Schauer wrote:
> > Package: graphviz
> > Version: 2.38.0-10
> > Severity: important
> > Control: block 804296 by -1
> > 
> > Hi,
> > 
> > I ran into this problem when building my package botch on mipsel:
> [...]
> > The problem can be trigged by the most minimal input:
> > 
> > > $ echo -ne 'digraph G {A -> B}' | dot
> > > dot: emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed.
> > Aborted
> 
> I can reproduce this using graphviz 2.38.0-11, but not 2.38.0-10.
> 
> The most obvious difference between them is that -11 is compiled with
> gcc-5 and -10 is not. That could be related (I havn't done much
> investigation so far)?

Yes, this is due to gcc-5 and compiling emit.c with -fno-ipa-sra
workarounds the problem. This is therefore likely the same issue than
bug #804536 / upstream bug PR68273, though I haven't analyzed the
generated code in details.

Aurelien

-- 
Aurelien Jarno  GPG: 4096R/1DDD8C9B
aurel...@aurel32.net http://www.aurel32.net


signature.asc
Description: Digital signature

Bug#804297: graphviz: dot on mipsel fails with emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed

[James Cowgill]

Control: notfound -1 2.38.0-10
Control: found -1 2.38.0-11

Hi,

On Sat, 2015-11-07 at 03:52 +0100, Johannes Schauer wrote:
> Package: graphviz
> Version: 2.38.0-10
> Severity: important
> Control: block 804296 by -1
> 
> Hi,
> 
> I ran into this problem when building my package botch on mipsel:
[...]
> The problem can be trigged by the most minimal input:
> 
>   > $ echo -ne 'digraph G {A -> B}' | dot
>   > dot: emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed.
>   Aborted

I can reproduce this using graphviz 2.38.0-11, but not 2.38.0-10.

The most obvious difference between them is that -11 is compiled with
gcc-5 and -10 is not. That could be related (I havn't done much
investigation so far)?

James

signature.asc
Description: This is a digitally signed message part

Bug#804297: graphviz: dot on mipsel fails with emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed

[Johannes Schauer]

Package: graphviz
Version: 2.38.0-10
Severity: important
Control: block 804296 by -1

Hi,

I ran into this problem when building my package botch on mipsel:

https://buildd.debian.org/status/fetch.php?pkg=botch=mipsel=0.17-2=1446862050

dot -T png buildgraph1.dot > buildgraph1.png
dot: emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed.
Aborted
Makefile:14: recipe for target 'buildgraph1.png' failed
make[2]: *** [buildgraph1.png] Error 134
make[2]: Leaving directory '/«PKGBUILDDIR»/doc/wiki'
Makefile:89: recipe for target 'doc' failed
make[1]: *** [doc] Error 2
make[1]: Leaving directory '/«PKGBUILDDIR»'
dh_auto_build: make -j1 returned exit code 2
debian/rules:6: recipe for target 'build-arch' failed
make: *** [build-arch] Error 2

botch uses dot to generate a png from a dot file which fails with:

dot: emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed.

but also only on mipsel. The other architectures build just fine.

The problem can be trigged by the most minimal input:

$ echo -ne 'digraph G {A -> B}' | dot
dot: emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed.
Aborted

To reproduce this problem, on a porter box do the following:

$ ssh etler.debian.org
josch@etler:~$ mysid=sid$RANDOM
josch@etler:~$ schroot -b -c sid -n $mysid
josch@etler:~$ dd-schroot-cmd -c $mysid apt-get update
josch@etler:~$ dd-schroot-cmd -c $mysid apt-get upgrade
josch@etler:~$ dd-schroot-cmd -c $mysid apt-get install graphviz
josch@etler:~$ schroot -r -c $mysid
(sid_mipsel-dchroot)josch@etler:~$ echo -ne 'digraph G {A -> B}' | dot
dot: emit.c:3873: bezier_bb: Assertion `bz.size > 0' failed.
Aborted
(sid_mipsel-dchroot)josch@etler:~$ logout
josch@etler:~$ schroot -e -c $mysid
josch@etler:~$ logout

Thanks!

cheers, josch