__emutls_get_address link problem

Simon Marlow marlowsd at gmail.com
Wed Aug 25 11:39:43 EDT 2010

On 25/08/2010 16:37, Christian Maeder wrote:
> Christian Maeder schrieb:
>> Simon Marlow schrieb:
>>> On 23/08/2010 17:45, Brandon S Allbery KF8NH wrote:
>>>> Hash: SHA1
>>>> On 8/23/10 11:57 , Christian Maeder wrote:
>>>>> However, when I try to compile the simplest source with on older
>>>>> gcc-3.4.4 I get the link error below, but only for the threaded rts!
>>>>> With ghc-6.12.1 and gcc-4.x.y (x<   3) I did not have such a problem when
>>>>> switching to a machine that only has gcc-3.4.4.
>>>> You got lucky; mixing GCC major versions like that is not expected (much
>>>> less guaranteed) to work in general.  GCC evidently changed the way it
>>>> implements some thread support routines between gcc3 and gcc4.
> Why should it not be possible to create object files (like also dynamic
> libs) with (very) different compilers?

This is a case where the ABI changed between versions of gcc (I 
presume).  It rarely happens with C code, but happens more often with 
C++, and, of course, in GHC we change the ABI with every single release 
so you should be used to this by now :-)

>>> In particular, thread-local variables, which are used by the parallel GC.
>> I'm just testing if I get runtime problems, but the linking error goes
>> away, if I put libgcc_eh.a from gcc-4.3.3 into ghc's libdir.
> The hack seems to work, although it looks dangerous regarding other
> symbols from libgcc_eh.a that should be in sync with the used gcc-3.4.4.

I wouldn't trust it.  Why do you need to mix code compiled by two 
different gcc versions?


More information about the Glasgow-haskell-users mailing list