Re: math/R: Build failure after PORTREVISION for shlib change

[David Wolfskill]

On Thu, Jun 29, 2017 at 12:39:50PM -0300, Joseph Mingrone wrote:
> ...
> With those options, the build succeeds in an 11.0-RELEASE-p1 jail.
> http://pkg.awarnach.mathstat.dal.ca/data/11amd64-default/2017-06-28_16h53m48s/logs/R-3.4.0_2.log
> ... 
> Everything matches.  The successful build in poudriere even has the same
> line that generates the error for you.
> 
> /usr/local/bin/gcc-ar5 -cr libtre.a regcomp.o regerror.o regexec.o tre-ast.o 
> tre-compile.o tre-match-approx.o tre-match-backtrack.o tre-match-parallel.o 
> tre-mem.o tre-parse.o tre-stack.o xmalloc.o
> 
> Could you have something in your environment or in make.conf that's
> causing a problem?
> 

As expected, further attempts to build math/R on my laptop have
continued to fail as before -- and as mentioned, I ran a poudriere run,
where building the package for math/R worked.

That said, the line in the poudriere log that corresponds to the failing
invocation is a bit different in my case.  (I found a couple of small
options differences, such as paper size; I don't believe those have
anything to do with the observed differences, but I've corrected them
for tomorrow's poudriere run.)

The line in the failing case (with the immediately preceding line
for context):

...
gcc5 -I. -I. -I../../../src/include -I../../../src/include -DLIBICONV_PLUG 
-I/usr/local/include -isystem /usr/local/include -DHAVE_CONFIG_H   -fopenmp 
-fpic  -O2 -pipe  -DLIBICONV_PLUG -fstack-protector 
-Wl,-rpath=/usr/local/lib/gcc5 -isystem /usr/local/include -fno-strict-aliasing 
-flto -fvisibility=hidden -c xmalloc.c -o xmalloc.o
/usr/local/bin/gcc-ar5 -cr libtre.a regcomp.o regerror.o regexec.o tre-ast.o 
tre-compile.o tre-match-approx.o tre-match-backtrack.o tre-match-parallel.o 
tre-mem.o tre-parse.o tre-stack.o xmalloc.o
ar: unrecognized option `--plugin'


and in the successful poudriere run:

...
cc -I. -I. -I../../../src/include -I../../../src/include -DLIBICONV_PLUG 
-I/usr/local/include -isystem /usr/local/include -DHAVE_CONFIG_H-fpic  -O2 
-pipe  -DLIBICONV_PLUG -fstack-protector -isystem /usr/local/include 
-fno-strict-aliasing  -fvisibility=hidden -c xmalloc.c -o xmalloc.o
/usr/local/bin/ar -cr libtre.a regcomp.o regerror.o regexec.o tre-ast.o 
tre-compile.o tre-match-approx.o tre-match-backtrack.o tre-match-parallel.o 
tre-mem.o tre-parse.o tre-stack.o xmalloc.o


Looking (on the laptop) at plausible "ar" executables in /usr/local/bin,
I see:

g1-227(12.0-C)[3] (cd /usr/local/bin && ls -lTi gcc-ar* ar)
1850562 -r-xr-xr-x  2 root  wheel  6368928 Mar 23 05:10:40 2017 ar
1850780 -r-xr-xr-x  2 root  wheel25728 May  1 05:47:28 2017 gcc-ar49
1846261 -r-xr-xr-x  2 root  wheel26192 Jun 30 04:57:30 2017 gcc-ar5
g1-227(12.0-C)[4] 

Apparently "ar" was installed by binutils-2.28,1, while the gcc-ar*
were installed by gcc49-4.9.4_3 and gcc5-5.4.0_2, respectively.

Weird.

Peace,
david
-- 
David H. Wolfskill  da...@catwhisker.org
Congratulations on the sesquicentennial, Canada! :-)

See http://www.catwhisker.org/~david/publickey.gpg for my public key.


signature.asc
Description: PGP signature

Re: math/R: Build failure after PORTREVISION for shlib change

[David Wolfskill]

On Thu, Jun 29, 2017 at 12:39:50PM -0300, Joseph Mingrone wrote:
> ...
> Everything matches.  The successful build in poudriere even has the same
> line that generates the error for you.

I may try ktracing the build

> /usr/local/bin/gcc-ar5 -cr libtre.a regcomp.o regerror.o regexec.o tre-ast.o 
> tre-compile.o tre-match-approx.o tre-match-backtrack.o tre-match-parallel.o 
> tre-mem.o tre-parse.o tre-stack.o xmalloc.o
> 
> Could you have something in your environment or in make.conf that's
> causing a problem?

g1-227(11.1)[1] grep -v '^#' /etc/make.conf
NET_SNMP_SYS_CONTACT="da...@catwhisker.org"
NET_SNMP_SYS_LOCATION="variable"
NET_SNMP_LOGFILE=/var/log/snmpd.log
NET_SNMP_PERSISTENTDIR=/var/net-snmp
SENDMAIL_MC=/etc/mail/laptop.mc
WITH_BSD_JDK=TRUE
WITHOUT_RUNTIME_CPUDETECTION=   YES
WITHOUT_CJK=YES
NO_SUID_XSERVER=YES
DEFAULT_VERSIONS+=linux=c6
INSTALL_AS_NCFTP=yes
OPTIONS_SET=OPTIMIZED_CFLAGS
DEFAULT_VERSIONS+=  perl5=5.24
FORCE_PKG_REGISTER= YES
PKG_NOCOMPRESS=1
g1-227(11.1)[2] 

> Joseph

I'll be trying a poudriere run Saturday morning, and will report
anything "interesting."

Thanks!

Peace,
david
-- 
David H. Wolfskill  da...@catwhisker.org
Do Congress's recent actions on health care qualify as "terrorist acts?"

See http://www.catwhisker.org/~david/publickey.gpg for my public key.


signature.asc
Description: PGP signature

Re: math/R: Build failure after PORTREVISION for shlib change

[Joseph Mingrone]

David Wolfskill  writes:
> g1-227(11.1)[2] make -C /usr/ports/math/R showconfig
> ===> The following configuration options are available for R-3.4.0_1:
>  ICU=on: Unicode support via ICU
>  INFO=on: GNU info manuals
>  LDOUBLE=on: Long double data type
>  LETTER=on: US letter paper
>  LIBR=on: Shared R library
>  MEMPROF=off: Memory profiling via Rprofmem() and tracemem()
>  NLS=on: Native Language Support
>  RPROF=on: R profiling via Rprof()
>  X11=on: X11 graphics device
> > Require GCC
>  LTO=on: Use Link Time Optimization
>  OPENMP=on: Parallel processing support via OpenMP
> > Require X11
>  GHOSTSCRIPT=on: Graphics device for bitmap files via Ghostscript
>  JPEG=on: JPEG graphics device
>  CAIROPANGO=on: Cairo graphics device and Pango multi-language text
>  PNG=on: PNG graphics device
>  TCLTK=on: Tcl/Tk GUI toolkit support
>  TEXDOCS=on: Build/Install TeX-dependent documentation files
>  TIFF=on: TIFF image format support
> > Options available for the single BLAS: you have to select exactly one 
> of them
>  ATLAS=off: ATLAS BLAS implementation
>  OPENBLAS=off: OpenBLAS BLAS implementation
>  NETLIB=off: Netlib BLAS implementation
>  RBLAS=on: Use R-bundled BLAS implementation
> ===> Use 'make config' to modify these settings
> g1-227(11.1)[3] 

With those options, the build succeeds in an 11.0-RELEASE-p1 jail.
http://pkg.awarnach.mathstat.dal.ca/data/11amd64-default/2017-06-28_16h53m48s/logs/R-3.4.0_2.log

> Just in case it's relevant, as /usr/local/bin/gcc-ar5 is from gcc5-5.4.0_2:

> g1-227(11.1)[5] make -C /usr/ports/lang/gcc5 showconfig
> ===> The following configuration options are available for gcc5-5.4.0_2:
>  BOOTSTRAP=on: Build using a full bootstrap
>  GRAPHITE=off: Support for Graphite loop optimizations
>  JAVA=on: Java platform support
> ===> Use 'make config' to modify these settings


Everything matches.  The successful build in poudriere even has the same
line that generates the error for you.

/usr/local/bin/gcc-ar5 -cr libtre.a regcomp.o regerror.o regexec.o tre-ast.o 
tre-compile.o tre-match-approx.o tre-match-backtrack.o tre-match-parallel.o 
tre-mem.o tre-parse.o tre-stack.o xmalloc.o

Could you have something in your environment or in make.conf that's
causing a problem?

Joseph


signature.asc
Description: PGP signature

Re: math/R: Build failure after PORTREVISION for shlib change

[Joseph Mingrone]

Hi David,

So far I haven't been able to reproduce any problems in an
11.0-RELEASE-p1 jail.  Could you send your list of port options?  If I
can't reproduce any issues with the same options turned on, I will ask
you for a full build log.

Joseph


signature.asc
Description: PGP signature

Re: math/R: Build failure after PORTREVISION for shlib change

[David Wolfskill]

On Wed, Jun 28, 2017 at 04:36:18PM -0300, Joseph Mingrone wrote:
> Hi David,
> 
> So far I haven't been able to reproduce any problems in an
> 11.0-RELEASE-p1 jail.  Could you send your list of port options?  If I
> can't reproduce any issues with the same options turned on, I will ask
> you for a full build log.
> 
> Joseph

Thanks for taking a look.

g1-227(11.1)[2] make -C /usr/ports/math/R showconfig
===> The following configuration options are available for R-3.4.0_1:
 ICU=on: Unicode support via ICU
 INFO=on: GNU info manuals
 LDOUBLE=on: Long double data type
 LETTER=on: US letter paper
 LIBR=on: Shared R library
 MEMPROF=off: Memory profiling via Rprofmem() and tracemem()
 NLS=on: Native Language Support
 RPROF=on: R profiling via Rprof()
 X11=on: X11 graphics device
> Require GCC
 LTO=on: Use Link Time Optimization
 OPENMP=on: Parallel processing support via OpenMP
> Require X11
 GHOSTSCRIPT=on: Graphics device for bitmap files via Ghostscript
 JPEG=on: JPEG graphics device
 CAIROPANGO=on: Cairo graphics device and Pango multi-language text
 PNG=on: PNG graphics device
 TCLTK=on: Tcl/Tk GUI toolkit support
 TEXDOCS=on: Build/Install TeX-dependent documentation files
 TIFF=on: TIFF image format support
> Options available for the single BLAS: you have to select exactly one of 
them
 ATLAS=off: ATLAS BLAS implementation
 OPENBLAS=off: OpenBLAS BLAS implementation
 NETLIB=off: Netlib BLAS implementation
 RBLAS=on: Use R-bundled BLAS implementation
===> Use 'make config' to modify these settings
g1-227(11.1)[3] 

I also placed both "normal" and gzipped copies of the typescript
from the (multiple) portmaster runs in
; see R.log (or
R.log.gz).

(I had run portmaster; it built a few things, tried math/R & died.
I re-started portmaster, eliding math/R, but portmaster found it
anyway... though it didn't show up in the list until several other
ports were built successfully.  I then tried again, skipping both
math/R and databases/R-cran-DBI; all of the specified ports built
OK.  Then I tried just math/R & databases/R-cran-DBI; those failed
(as before).  I then specified 'MAKE_JOBS_UNSAFE=yes' in /etc/make.conf
& tried the laast run again; still failed.  Each time, I appended
to the typescript.)

Just in case it's relevant, as /usr/local/bin/gcc-ar5 is from gcc5-5.4.0_2:

g1-227(11.1)[5] make -C /usr/ports/lang/gcc5 showconfig
===> The following configuration options are available for gcc5-5.4.0_2:
 BOOTSTRAP=on: Build using a full bootstrap
 GRAPHITE=off: Support for Graphite loop optimizations
 JAVA=on: Java platform support
===> Use 'make config' to modify these settings

Peace,
david
-- 
David H. Wolfskill  da...@catwhisker.org
Do Congress's recent actions on health care qualify as "terrorist acts?"

See http://www.catwhisker.org/~david/publickey.gpg for my public key.


signature.asc
Description: PGP signature