Opened 10 years ago

Closed 10 years ago

Last modified 9 years ago

#45483 closed defect (fixed)

gcc49: bug in libstdc++ when using std::complex

Reported by: vsoftco@… Owned by: larryv (Lawrence Velázquez)
Priority: Normal Milestone:
Component: ports Version: 2.3.2
Keywords: Cc: mww@…
Port: gcc49

Description

The following produces a segfault when compiling with g++4.9.1 using libstdc++ that comes with it.

#include <complex>
#include <iostream>

int main()
{
	std::complex<double> a{1,0};
	std::cout << a.imag() << std::endl; // works fine
	std::cout << a << std::endl; // segfault
}

Probably there is something related to ostream operator<< overloading for std::complex, although I cannot say for sure. However, the bug is extremely important, as it makes the whole std::complex part of the library unusable.

Attachments (1)

yosemite-symbol-lookup.patch (11.6 KB) - added by larryv (Lawrence Velázquez) 10 years ago.

Download all attachments as: .zip

Change History (12)

comment:1 Changed 10 years ago by ryandesign (Ryan Carsten Schmidt)

Keywords: gcc libstdc++ removed
Owner: changed from macports-tickets@… to mww@…
Priority: HighNormal
Summary: bug in libstdc++ when using std::complexgcc49: bug in libstdc++ when using std::complex

You should probably test if the problem remains in gcc5, and also report the problem to the developers of gcc. All we do at MacPorts is package the software they provide.

comment:2 Changed 10 years ago by larryv (Lawrence Velázquez)

Cc: larryv@… added

I can reproduce the segfault.

comment:3 in reply to:  1 ; Changed 10 years ago by vsoftco@…

Replying to ryandesign@…:

You should probably test if the problem remains in gcc5, and also report the problem to the developers of gcc. All we do at MacPorts is package the software they provide.

I tested the code with gcc4.9.1 from Homebrew (and using its libstdc++), and there is no segfault. I have also filled a bug with gcc, however it seems extremely strange that one version of gcc4.9.1 works and another one not.

Changed 10 years ago by larryv (Lawrence Velázquez)

comment:4 in reply to:  3 ; Changed 10 years ago by larryv (Lawrence Velázquez)

I know what’s going on. The crash log looks uncomfortably familiar…

Exception Type:        EXC_BAD_ACCESS (SIGSEGV)
Exception Codes:       EXC_I386_GPFLT

Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0   libdyld.dylib                 	0x00007fff82b01432 stack_not_16_byte_aligned_error + 0

…because this is my old friend, the Libtool MACOSX_DEPLOYMENT_TARGET bug.

I’ve attached a preliminary patch. Could you try it?

% sudo port uninstall libgcc gcc49 && sudo port clean libgcc gcc49
% sudo port patch libgcc
% curl https://trac.macports.org/raw-attachment/ticket/45483/yosemite-symbol-lookup.patch | sudo patch -p1 -d $(port work libgcc)/gcc-4.9.1
% sudo port install libgcc
% sudo port patch gcc49
% curl https://trac.macports.org/raw-attachment/ticket/45483/yosemite-symbol-lookup.patch | sudo patch -p1 -d $(port work libgcc)/gcc-4.9.1
% sudo port install gcc49
Last edited 10 years ago by larryv (Lawrence Velázquez) (previous) (diff)

comment:5 in reply to:  4 Changed 10 years ago by vsoftco@…

Thanks, will try later today and let you know.

comment:6 Changed 10 years ago by larryv (Lawrence Velázquez)

Cc: mww@… added; larryv@… removed
Owner: changed from mww@… to larryv@…

I’ve pushed the change in r127092. Can you run selfupdate and try the new version (4.9.1_1)?

comment:7 Changed 10 years ago by larryv (Lawrence Velázquez)

Resolution: fixed
Status: newclosed

Fixed gcc48 in r127093 and gcc5 in r127094.

comment:8 Changed 10 years ago by vsoftco@…

It works, many thanks for the super fast fix!

comment:9 Changed 10 years ago by vsoftco@…

I have a question, is the bug related to libgcc or to macports? If it's related to macports, I'll remove the bug I filled for g++. Thanks again for the great work in fixing it!

comment:10 in reply to:  9 Changed 10 years ago by larryv (Lawrence Velázquez)

It’s an upstream bug. The configure scripts in the distribution are based on the current Libtool, which uses incorrect linker flags when MACOSX_DEPLOYMENT_TARGET is set to “10.10”. We set that environment variable in our builds; Homebrew does not, which is why you didn’t observe the bug there.

Could you provide a link to the bug report you filed?

comment:11 Changed 9 years ago by larryv (Lawrence Velázquez)

Note: See TracTickets for help on using tickets.