octave-devel: error: 'CGDisplayBitsPerPixel' was not declared in this scope

[skymoo (Adam Mercer)]

Looks like the #30737 also effects Mavericks as I'm getting the following:

:info:build libtool: compile:  /opt/local/bin/g++-mp-4.8 -DHAVE_CONFIG_H -I. -I.. -I../libgnu -I../libgnu -I../libcruft/misc -I../liboctave -I../liboctave -I. -I. -pipe -Os -m64 -DHAVE_CONFIG_H -I/opt/local/include -Wall -W -Wshadow -Wold-style-cast -Wformat -Wpointer-arith -Wwrite-strings -Wcast-align -Wcast-qual -pipe -Os -m64 -D_THREAD_SAFE -pthread -pipe -Os -m64 -c dynamic-ld.cc  -fno-common -DPIC -o .libs/liboctinterp_la-dynamic-ld.o
:info:build display.cc: In member function 'void display_info::init(bool)':
:info:build display.cc:75:46: error: 'CGDisplayBitsPerPixel' was not declared in this scope
:info:build            dp = CGDisplayBitsPerPixel (display);
:info:build                                               ^

Using the same fix as #30737, i.e.:

platform darwin 13 {
    # In 10.9 patch is required as above, also
    # the LANG environment variable needs to
    # be set to "C" otherwise /usr/bin/sed
    # fails with an error, if you installed gsed
    # with default name this should have no effect.

    patchfiles-append patch-src-display.cc.diff
    build.args-append LANG="C"
}

allows the build to get a little further but then it fails with:

Undefined symbols for architecture x86_64:
  "Magick::CoderInfo::CoderInfo(std::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)", referenced from:
      F__magick_format_list__(octave_value_list const&, int) in DLD_FUNCTIONS___magick_read___la-__magick_read__.o
ld: symbol(s) not found for architecture x86_64

full build log is attached.

[michaelld (Michael Dickens)]

I don't have 10.9 to test on, so this might have to wait for a bit. In the meantime, does the following command:

nm -a /opt/local/lib/libGraphicsMagick++.dylib | grep CoderInfo

return something like "__ZN6Magick9CoderInfoC1ERKSs" or "__ZN6Magick9CoderInfoC2ERKSs" (without the "!" which the wiki requires to get the formatting correct) as one item?

[skymoo (Adam Mercer)]

Replying to michaelld@…:

I don't have 10.9 to test on, so this might have to wait for a bit.

No rush.

In the meantime, does the following command:

nm -a /opt/local/lib/libGraphicsMagick++.dylib | grep CoderInfo

return something like "__ZN6Magick9CoderInfoC1ERKSs" or "__ZN6Magick9CoderInfoC2ERKSs" (without the "!" which the wiki requires to get the formatting correct) as one item?

Doesn't look like it:

[ram@cizin ~]$ nm -a /opt/local/lib/libGraphicsMagick++.dylib | grep CoderInfo
0000000000005abc T __ZN6Magick9CoderInfoC1EPKN9MagickLib11_MagickInfoE
0000000000005844 T __ZN6Magick9CoderInfoC1ERKNSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEE
00000000000057a2 T __ZN6Magick9CoderInfoC1ERKS0_
0000000000005722 T __ZN6Magick9CoderInfoC1Ev
0000000000005ac6 T __ZN6Magick9CoderInfoC2EPKN9MagickLib11_MagickInfoE
000000000000584e T __ZN6Magick9CoderInfoC2ERKNSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEE
00000000000057ac T __ZN6Magick9CoderInfoC2ERKS0_
0000000000005762 T __ZN6Magick9CoderInfoC2Ev
00000000000059de T __ZN6Magick9CoderInfoD1Ev
00000000000059e8 T __ZN6Magick9CoderInfoD2Ev
0000000000005a74 T __ZN6Magick9CoderInfoaSERKS0_
0000000000005a56 T __ZNK6Magick9CoderInfo10isReadableEv
0000000000005a60 T __ZNK6Magick9CoderInfo10isWritableEv
0000000000005a3a T __ZNK6Magick9CoderInfo11descriptionEv
0000000000005a6a T __ZNK6Magick9CoderInfo12isMultiFrameEv
0000000000005a22 T __ZNK6Magick9CoderInfo4nameEv
[ram@cizin ~]$

[michaelld (Michael Dickens)]

The entry

__ZN6Magick9CoderInfoC1ERKNSt3__112basic_stringIcNS1_11char_traitsIcEENS1_9allocatorIcEEEE

is what we're looking for, but it seems incorrect. It demangles to (for me, running 10.8):

Magick::CoderInfo::CoderInfo(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)

with the extra "__1::" (without the "!") compared with the error you got. So, either Octave or GraphicsMagick is messing with std::string. Looking at the code for GraphicsMagick++ (GraphicsMagick-1.3.18/Magick++/lib/CoderInfo.cpp:37) reads

Magick::CoderInfo::CoderInfo ( const std::string &name_ )

so, I'd guess that it is your install of GraphicsMagick that is the culprit here, not Octave.

One way to test this is to create a simple program which uses std::string, compile and link it, then do an "nm -a" on the resulting binary and see which mangled C++ expression comes up (using std::basic_string or std::__1::basic_string).

[skymoo (Adam Mercer)]

Replying to michaelld@…:

One way to test this is to create a simple program which uses std::string, compile and link it, then do an "nm -a" on the resulting binary and see which mangled C++ expression comes up (using std::basic_string or std::__1::basic_string).

Do you have a simple code example, as I know next to nothing about C++?

[michaelld (Michael Dickens)]

Sure; save the following as "test_std_string.cc":

#include <iostream>

void print (const std::string& str)
{ std::cout << str << std::endl; }

int main (void)
{ print ("foo"); return (1); }

then compile as:

clang++ -o test_std_string -pipe -g test_std_string.cc

and, if that succeeds,

c++filt `nm -a test_std_string | grep print | grep " T " | awk '{ print $NF }'`

should find, demangle, and print out the actual function. When I do the above, I get back:

print(std::string const&)

as expected.

[skymoo (Adam Mercer)]

Something very different here:

[ram@cizin ~]$ cat test_std_string.cc
#include <iostream>

void print (const std::string& str)
{ std::cout << str << std::endl; }

int main (void)
{ print ("foo"); return (1); }
[ram@cizin ~]$ clang++ -o test_std_string -pipe -g test_std_string.cc
[ram@cizin ~]$ c++filt `nm -a test_std_string | grep print | grep " T " | awk '{ print $NF }'`
print(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)
[ram@cizin ~]$

[michaelld (Michael Dickens)]

That's strange, but I guess that's the way it is. So, the issue is probably with Octave after all. I'll push this bug report upstream and see what they say.

[skymoo (Adam Mercer)]

Thanks, let me know if I can help with anything.

[bpabbott@…]

I don't have a 10.9 box to test on, but we've applied a fix to display.cc related to the CGDisplayBitsPerPixel problem. The link to the patch is below.

​http://hg.savannah.gnu.org/hgweb/octave/diff/0eb1b1eb2c76/src/display.cc

[jsalort (Julien Salort)]

Cc Me!

[michaelld (Michael Dickens)]

I've finally downloaded 10.9, but I'm busy through this weekend. Hopefully I'll find time next week to work on this ticket.

[AP1010]

Replying to michaelld@…:

I've finally downloaded 10.9, but I'm busy through this weekend. Hopefully I'll find time next week to work on this ticket.

FWIW: I managed to successfully build octave-devel on Mavericks. After applying the "platform darwin 13" patch, I fixed the GraphicsMagick issue by building it with gcc48 (not default clang).

[jasonxh@…]

Cc Me!

[macports@…]

Cc Me!

[roberthuston@…]

Cc Me!

[roberthuston@…]

Replying to arthur@…:

Replying to michaelld@…:

I've finally downloaded 10.9, but I'm busy through this weekend. Hopefully I'll find time next week to work on this ticket.

FWIW: I managed to successfully build octave-devel on Mavericks. After applying the "platform darwin 13" patch, I fixed the GraphicsMagick issue by building it with gcc48 (not default clang).

I applied the "platform darwin 13" patch also, but I still get the same error. I uninstalled and cleaned GraphicsMagick and then did a port select gcc to "mp-gcc48" before reinstalling GraphicsMagick. Do I have to do something different to build it with gcc48?

[roberthuston@…]

Replying to roberthuston@…:

Replying to arthur@…:

Replying to michaelld@…:

I've finally downloaded 10.9, but I'm busy through this weekend. Hopefully I'll find time next week to work on this ticket.

FWIW: I managed to successfully build octave-devel on Mavericks. After applying the "platform darwin 13" patch, I fixed the GraphicsMagick issue by building it with gcc48 (not default clang).

I applied the "platform darwin 13" patch also, but I still get the same error. I uninstalled and cleaned GraphicsMagick and then did a port select gcc to "mp-gcc48" before reinstalling GraphicsMagick. Do I have to do something different to build it with gcc48?

Never mind, I got it to work. Used

  sudo port install GraphicsMagick configure.compiler=macports-gcc-4.8

and then re-ran

  sudo port install octave-devel +atlas +gcc48

[michaelld (Michael Dickens)]

I'm running a 10.9 boot disk, currently downloaded xcode. I'll hopefully be able to verify all of the above in ~2 hours. Plenty of fixes for 10.9 needed in my ports :)

[michaelbartz@…]

Cc Me!

[lawrence.ong@…]

Cc Me!

[lpsinger (Leo Singer)]

Cc Me!

[davydden@…]

Cc Me!

[larryv (Lawrence Velázquez)]

You have to click the “CcMe!” button next to the Cc list above to actually Cc yourself. I have done this.

[ryandesign (Ryan Carsten Schmidt)]

Has duplicates #40930, #41059, #41091.

[onurdomanic@…]

Cc Me!

[onurdomanic@…]

Cc Me!

[iqgrande]

Cc Me!

[iqgrande]

Cc Me!

[jendrek@…]

Replying to ram@…:

Something very different here:

[ram@cizin ~]$ clang++ -o test_std_string -pipe -g test_std_string.cc
[ram@cizin ~]$ c++filt `nm -a test_std_string | grep print | grep " T " | awk '{ print $NF }'`
print(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)
[ram@cizin ~]$

Same here...

[jendrek@…]

Replying to jendrek@…:

Replying to ram@…:

Something very different here:

[ram@cizin ~]$ clang++ -o test_std_string -pipe -g test_std_string.cc
[ram@cizin ~]$ c++filt `nm -a test_std_string | grep print | grep " T " | awk '{ print $NF }'`
print(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&)
[ram@cizin ~]$

Same here...

Except that my ports are 2.2.1... not 2.2.0 as listed in this ticket... 2.2.1 installed from scratch...

[ralph@…]

Cc Me!

[michaelld (Michael Dickens)]

I'm up to getting atlas installed, which is taking just !@#$ forever; then metis, suitesparse, and maybe a few others which should be fast in comparison. I should be working on this ticket either later today or tomorrow.

[shauriege@…]

Cc Me!

[lukasz.m.dobrzanski@…]

Cc Me!

[apple@…]

Cc Me!

[michaelld (Michael Dickens)]

For folks having this issue, I think the following should work:

sudo port -f uninstall GraphicsMagick
sudo port install GraphicsMagick configure.compiler=macports-gcc-4.8
sudo port clean octave-devel
sudo port install octave-devel +gcc48 configure.compiler=macports-gcc-4.8

I'm working on a clang/clang solution too.

[lord@…]

Michael, I tried you suggestion and still getting the same as before build error:

:info:build display.cc:75:46: error: 'CGDisplayBitsPerPixel' was not declared in
 this scope

[shauriege@…]

Replying to lord@…:

Michael, I tried you suggestion and still getting the same as before build error:

:info:build display.cc:75:46: error: 'CGDisplayBitsPerPixel' was not declared in
 this scope

Michael's fix works for me when I apply the Darwin 13 fix discussed in the ticket description above.

[michaelld (Michael Dickens)]

Sorry; I needed to push my changed Portfile, which I just did in r112810. At least for me on 10.9, octave-devel now works with [GraphicsMagick/octave-devel] compiled with clang/clang or gcc/gcc; I have not tried combining. Meaning: putting configure.compiler=macports-gcc-4.8 after the port command. The +gcc48 is for Fortran only now in octave-devel. This change should be live by ~5:10 PM/US/ET, if not sooner. You'll need to do:

sudo port clean octave-devel
sudo port selfupdate
sudo port install octave-devel

or whatever your special sauce is, and hopefully octave-devel will work now on 10.9 (as well as continue to work on 10.8 or older). I'm leaving this ticket open until either I hear confirmation or a week has passed without a peep.

[roberthuston@…]

Replying to michaelld@…:

Sorry; I needed to push my changed Portfile, which I just did in r112810. At least for me on 10.9, octave-devel now works with [GraphicsMagick/octave-devel] compiled with clang/clang or gcc/gcc; I have not tried combining. Meaning: putting configure.compiler=macports-gcc-4.8 after the port command. The +gcc48 is for Fortran only now in octave-devel. This change should be live by ~5:10 PM/US/ET, if not sooner. You'll need to do:

sudo port clean octave-devel
sudo port selfupdate
sudo port install octave-devel

or whatever your special sauce is, and hopefully octave-devel will work now on 10.9 (as well as continue to work on 10.8 or older). I'm leaving this ticket open until either I hear confirmation or a week has passed without a peep.

I still get failures in building octave-devel even after performing the above steps. I uninstalled (and cleaned) GraphicsMagick, and then all of my octave packages, and finally octave-revel itself. I then did a selfupdate, updated my outdated ports (there were a few) and then did a "sudo port install octave-devel +atlas +gcc48." I ended up rebuilding both GraphicsMagick and octave-devel with the "configure.compiler=macports-gcc4.8" option to get things to work again.

[damian@…]

Cc Me!

[jendrek@…]

Replying to michaelld@…: Sorry; still problems... I am attaching the log.

[michaelld (Michael Dickens)]

Same issue for both of you: clang is being picky about class instantiation when checking for existence. Can you also post the file "port work octave-devel/octave-3.6.4/config.log"? I thought my patch to configure takes care of this issue, but maybe not.

[michaelld (Michael Dickens)]

My bad. I forgot to add the patch to the patchfiles. Should now be fixed in r112833.

[jendrek@…]

Replying to michaelld@…:

My bad. I forgot to add the patch to the patchfiles. Should now be fixed in r112833.

Still fails... I selfupdated, cleaned, and re-run the build...

Am attaching the log.

[michaelbartz@…]

I did not get too far:

Error: Unable to open port: extra characters after close-quote
Error: Unable to execute port: upgrade qhull failed

[michaelld (Michael Dickens)]

@michaelbartz: qhull uses CMake, and the CMake PortGroup has been in flux today. Hopefully it's stable again.

@jendrek: The log implies to me that the patch is not in place; not your fault; probably not enough time between when I posted and when you updated: I've been told it takes ~30 minutes for changes to be "live".

So, both of you, try this yet again -- we -will- get there sooner or later :)

sudo port clean octave-devel
sudo port selfupdate
sudo port install octave-devel

Sometimes 'port' just requires a few good cleans and retries to get working ...

[michaelbartz@…]

I can confirm that octave-devel complies successfully now.

[jendrek@…]

I can also compile now.

Thank you very much!

[lukasz.m.dobrzanski@…]

x3 should be enough to mark as fixed (octave-devel complies successfully now). Thank you Michael.

[roberthuston@…]

FWIW: I can build it now too. Thank you so very much for fixing this!

[michaelld (Michael Dickens)]

Excellent! Y'all are welcome! I'm closing this ticket as fixed.