www.digitalmars.com         C & C++   DMDScript  

D.gnu - Unable to compile gdc with gcc 4.8.2 on OSX 10.8.5

reply "kahuna" <kahuna0k gmail.com> writes:
I've been unable to compile gcc patches to gcc-4.8 on OSX, I get 
an error in thread.d, an assert "Not implemented", it seems that 
it is not getting the right version while compiling initStack() 
and fiber_entryPoint(), if I comment out both asserts, as 
expected it breaks the library and whenever I compile anything 
(even an empty main()) I get a lot of errors about undefined 
symbols similar to _D11TypeInfo_Am6__initZ in libgphobos2.a

It is gdc supported under OSX??
Oct 31 2013
parent reply "kahuna" <kahuna0k gmail.com> writes:
On Friday, 1 November 2013 at 06:08:53 UTC, kahuna wrote:

 I've been unable to compile gcc patches to gcc-4.8 on OSX, I 
 get an error in thread.d, an assert "Not implemented", it seems 
 that it is not getting the right version while compiling 
 initStack() and fiber_entryPoint(), if I comment out both 
 asserts, as expected it breaks the library and whenever I 
 compile anything (even an empty main()) I get a lot of errors 
 about undefined symbols similar to _D11TypeInfo_Am6__initZ in 
 libgphobos2.a

 It is gdc supported under OSX??


I have tried to compile branch v1.x with gcc-4.7.3 and although I 
get no error on thread.d (I get some bad references to direct 
d_name.ptr, fixed removing .ptr) but I get the same undefined 
symbols when compiling with gdc, so I understand that OSX is not 
currently working, at least on 10.8.5 and Xcode 5
Nov 01 2013
parent reply "Joseph Rushton Wakeling" <joseph.wakeling webdrake.net> writes:
On Friday, 1 November 2013 at 07:51:09 UTC, kahuna wrote:

 I have tried to compile branch v1.x with gcc-4.7.3 and although 
 I get no error on thread.d (I get some bad references to direct 
 d_name.ptr, fixed removing .ptr) but I get the same undefined 
 symbols when compiling with gdc, so I understand that OSX is 
 not currently working, at least on 10.8.5 and Xcode 5


Are you sure it's not because Xcode 5 has clang as the default 
compiler, even symlinking g++ to clang++? It may therefore be 
that gcc won't compile with clang, and installing GCC proper will 
solve your problem.
Nov 01 2013
next sibling parent "kahuna" <kahuna0k gmail.com> writes:
On Friday, 1 November 2013 at 10:03:42 UTC, Joseph Rushton 
Wakeling wrote:

 On Friday, 1 November 2013 at 07:51:09 UTC, kahuna wrote:

 I have tried to compile branch v1.x with gcc-4.7.3 and 
 although I get no error on thread.d (I get some bad references 
 to direct d_name.ptr, fixed removing .ptr) but I get the same 
 undefined symbols when compiling with gdc, so I understand 
 that OSX is not currently working, at least on 10.8.5 and 
 Xcode 5


 Are you sure it's not because Xcode 5 has clang as the default 
 compiler, even symlinking g++ to clang++? It may therefore be 
 that gcc won't compile with clang, and installing GCC proper 
 will solve your problem.


I have compiled it gcc-4.8.1 but anyway it should bootstrap 
itself, shouldn't it?
Nov 01 2013
prev sibling parent reply "kahuna" <kahuna0k gmail.com> writes:
On Friday, 1 November 2013 at 10:03:42 UTC, Joseph Rushton 
Wakeling wrote:

 On Friday, 1 November 2013 at 07:51:09 UTC, kahuna wrote:

 I have tried to compile branch v1.x with gcc-4.7.3 and 
 although I get no error on thread.d (I get some bad references 
 to direct d_name.ptr, fixed removing .ptr) but I get the same 
 undefined symbols when compiling with gdc, so I understand 
 that OSX is not currently working, at least on 10.8.5 and 
 Xcode 5


 Are you sure it's not because Xcode 5 has clang as the default 
 compiler, even symlinking g++ to clang++? It may therefore be 
 that gcc won't compile with clang, and installing GCC proper 
 will solve your problem.


I have compiled it with gcc-4.8.1, but anyway it should bootstrap 
itself.
Nov 01 2013
parent reply Joseph Rushton Wakeling <joseph.wakeling webdrake.net> writes:
On 02/11/13 04:56, kahuna wrote:

 I have compiled it with gcc-4.8.1, but anyway it should bootstrap itself.


Just to avoid doubt, I want to be absolutely clear what you're doing here.

When you say "with gcc-4.8.1" are you talking about the actual GCC compiler you 
already have installed on your system, or the GCC source that you're using with
GDC?

The point is that once those gcc sources are built, that will indeed bootstrap 
compilation of the rest of gcc and GDC; but you need a working compiler that
can 
compile the core of GCC itself, and it's unlikely clang can do that.  It's 
likely with Xcode 5 that the system compiler is clang and not gcc, and that the 
gcc/g++ commands are symlinked to clang.

Just to test, can you try typing g++ --version in a terminal window and see
what 
it outputs?

Hope that I don't sound condescending here, I'm just trying to make sure that I 
don't misunderstand you and that I understand properly how your system is set
up.
Nov 02 2013
parent reply "Joseph Rushton Wakeling" <joseph.wakeling webdrake.net> writes:
On Saturday, 2 November 2013 at 10:49:50 UTC, Joseph Rushton 
Wakeling wrote:

 On 02/11/13 04:56, kahuna wrote:

 I have compiled it with gcc-4.8.1, but anyway it should 
 bootstrap itself.


 Just to avoid doubt, I want to be absolutely clear what you're 
 doing here.


Kahuna, any news? Did you manage to get your build working?
Nov 02 2013
next sibling parent "kahuna" <kahuna0k gmail.com> writes:
On Sunday, 3 November 2013 at 05:51:07 UTC, Joseph Rushton 
Wakeling wrote:

 On Saturday, 2 November 2013 at 10:49:50 UTC, Joseph Rushton 
 Wakeling wrote:

 On 02/11/13 04:56, kahuna wrote:

 I have compiled it with gcc-4.8.1, but anyway it should 
 bootstrap itself.


 Just to avoid doubt, I want to be absolutely clear what you're 
 doing here.


 Kahuna, any news? Did you manage to get your build working?


Sorry I was without internet during the weekend (THE HELL ON 
EARTH!!! :-), I'm using gcc 4.8.1 binary distribution for osx to 
compile gcc-4.8.2, I also tried gcc-4.7.0 to compile latest 
gcc-4.7.3 with identical result (the gcc-4.8.1 was downloaded 
from 
http://sourceforge.net/projects/gnuada/files/GNAT_GCC%20Mac%20OS%20X/4.8.1/) 
and gcc-4.7.0 from the equivalent place for it.

the problem seems to be related to double underscore and simple 
underscore during the linking process. Compiling an empty main() 
and then linking produce lot of errors similar to

Undefined symbols for architecture x86_64:
   "_D10TypeInfo_a6__initZ", referenced from:
….

if I make

nm libgphobos.a | grep D10TypeInfo_a6__initZ

I get:

                  U _D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ
00000000000002b0 S __D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ

so maybe the compiler is failing when building the libgphobos.a

With gcc-4.8.2 I get the equivalent for libgphobos2.a

maybe the linker for OSX is not compatible with something deep 
inside gdc?? I have compiled c++ and c code without problems 
using those compilers and the references for c++ classes are 
correctly constructed with double underscore prefix..

Any idea?
Nov 04 2013
prev sibling parent reply "kahuna" <kahuna0k gmail.com> writes:
On Sunday, 3 November 2013 at 05:51:07 UTC, Joseph Rushton 
Wakeling wrote:

 On Saturday, 2 November 2013 at 10:49:50 UTC, Joseph Rushton 
 Wakeling wrote:

 On 02/11/13 04:56, kahuna wrote:

 I have compiled it with gcc-4.8.1, but anyway it should 
 bootstrap itself.


 Just to avoid doubt, I want to be absolutely clear what you're 
 doing here.


 Kahuna, any news? Did you manage to get your build working?


Sorry I was without internet during the weekend (THE HELL ON 
EARTH!!! :-), I'm using gcc 4.8.1 binary distribution for osx to 
compile gcc-4.8.2, I also tried gcc-4.7.0 to compile latest 
gcc-4.7.3 with identical result (the gcc-4.8.1 was downloaded 
from 
http://sourceforge.net/projects/gnuada/files/GNAT_GCC%20Mac%20OS%20X/4.8.1/) 
and gcc-4.7.0 from the equivalent place for it.

the problem seems to be related to double underscore and simple 
underscore during the linking process. Compiling an empty main() 
and then linking produce lot of errors similar to

Undefined symbols for architecture x86_64:
   "_D10TypeInfo_a6__initZ", referenced from:
….

if I make

nm libgphobos.a | grep D10TypeInfo_a6__initZ

I get:

                  U _D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ
00000000000002b0 S __D10TypeInfo_a6__initZ
                  U _D10TypeInfo_a6__initZ

so maybe the compiler is failing when building the libgphobos.a

With gcc-4.8.2 I get the equivalent for libgphobos2.a

maybe the linker for OSX is not compatible with something deep 
inside gdc?? I have compiled c++ and c code without problems 
using those compilers and the references for c++ classes are 
correctly constructed with double underscore prefix.

In fact after searching for this specific behavior I found this:

http://forum.dlang.org/thread/4DE7ABFD.5000808 cam.ac.uk

so it seems that this has been already reported a year ago 
without any fix since then :-/

Any idea?
Nov 04 2013
parent reply "kahuna" <kahuna0k gmail.com> writes:
More info on this,

It seems that latest version available from bitbucket do compile 
at least inside 4.7.3 without problems, I'll try to compile 
libgphobos2.a from this version using the new gdc from and check 
if it is a problem with gdc or with the libgphobos2 Makefiles.

Regards
Nov 06 2013
parent reply "Joseph Rushton Wakeling" <joseph.wakeling webdrake.net> writes:
On Thursday, 7 November 2013 at 06:07:53 UTC, kahuna wrote:

 It seems that latest version available from bitbucket


Are your previous build attempts also from BitBucket-hosted code? 
The most up-to-date GDC source is on GitHub.
Nov 06 2013
next sibling parent Iain Buclaw <ibuclaw ubuntu.com> writes:
On 7 November 2013 07:00, Joseph Rushton Wakeling <
joseph.wakeling webdrake.net> wrote:


 On Thursday, 7 November 2013 at 06:07:53 UTC, kahuna wrote:


 It seems that latest version available from bitbucket


 Are your previous build attempts also from BitBucket-hosted code? The most
 up-to-date GDC source is on GitHub.


See Type::getTypeInfoIdent where the frontend force changes the symbol
name.  Am yet to come to an agreement on how to deal with backend-specific
symbol prefixes.   C++ mangling is also likely broken too if you are seeing
this behaviour. (see cpp_mangle).

Regards
-- 
Iain Buclaw

*(p < e ? p++ : p) = (c & 0x0f) + '0';
Nov 07 2013
prev sibling parent "kahuna" <kahuna0k gmail.com> writes:
On Thursday, 7 November 2013 at 07:00:56 UTC, Joseph Rushton 
Wakeling wrote:

 On Thursday, 7 November 2013 at 06:07:53 UTC, kahuna wrote:

 It seems that latest version available from bitbucket


 Are your previous build attempts also from BitBucket-hosted 
 code? The most up-to-date GDC source is on GitHub.


No my previous failed builds are with GitHub code, I tried 
BitBucket to check if it was something happening since ever.
Nov 07 2013