Le samedi 09 octobre 2010 20:56:01, vous avez écrit :
> On Sat, 2010-10-09 at 19:47 +0200, Thomas Preud'homme wrote:
> > tcc in squeeze is unable to link a program into an executable on amd64
> > when it uses string functions (strlen and co). As almost all C program
> > use string functions, I consider this bug serious. The version in
> > unstable contains a fix and it consists of just a few lines. The fix
> > was introduced in tcc 0.9.25-4 which stayed in unstable for almost 2
> > months without bringing any new bug.
>
> +--- a/libtcc.c
> ++++ b/libtcc.c
> +@@ -1431,7 +1431,11 @@ static void rt_printline(unsigned long wanted_pc)
> [...]
> ++#ifdef STT_IFUNC
> ++ if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
>
> Should the ifdef not also refer to STT_GNU_IFUNC? This appears to be
> the only occurrence of STT_IFUNC in the package.
No it should not. In fact the ifdef should simply go away. But part of the
code is really not important. It's about displaying source line number in case
of runtime error and the function in which the error is. As the ifdef isn't
correct (it should use STT_GNU_IFUNC or simply not exists), and the function
where the error happen is a STT_GNU_IFUNC symbol, it won't display the
function name but just the PC. In other words, it's not a regression and as
only string functions in (e)glibc seems to use STT_GNU_IFUNC the effect is
limited. So, should I make a new upload in unstable?
>
> > Version 0.9.25-5 is there only for
> > a few days but the only change is to drop armel architecture as tcc
> > doesn't build on armel because of a failure in the test suite.
>
> p-a-s still says that tcc should be built on armel; that should also be
> fixed.
p-a-s bug submitted to buildd.debian.org (see #599674)
>
> Regards,
>
> Adam
Attachment:
signature.asc
Description: This is a digitally signed message part.