Bug#1144189: c-t-b-ports fails to building glibc for arc-linux-gnu

Hi,

First of all, I just realized that the linux kernel now provides a 
libc6-dev-arc-cross package in addition to linux-libc-dev. I got tricked 
by the fact that cross-toolchain-base-ports is able to use the multiarch 
headers, and it uses them since it stopped building the 
libc6-dev-$arch-cross packages. Except for the arc architectures as 
linux-libc-dev doesn't provide the headers for that architecture, hence 
the FTBFS.

Therefore let me answer you the rest of the mail with that in mind.

On 2026-08-13 21:08, Bastian Blank wrote:
> Hi
> 
> On Thu, Aug 13, 2026 at 08:40:15PM +0200, Aurelien Jarno wrote:
> > It does matter. Your solution works at build time, but it doesn't work 
> > at runtime.
> 
> What are you talking about?!?  runtime or build time does not make a
> difference on the existence of the kernel headers in
> /usr/GNU_HOST/include.  Please explain how this could happen.

I meant as runtime, like when using arc-linux-gnu-gcc with the 
libc-dev-arc-cross package installed. But you are correct that the cross 
compiler is also able to use the headers provided by the 
linux-libc-dev-arc-cross build from src:linux.

> >             Something has to provide the kernel headers for the cross 
> > glibc and cross compiler to work. In the past it was 
> > linux-libc-dev-arc-cross and libc6-dev-arc-cross depends on it for that 
> > purpose.
> 
> And what changed?  libc6-dev-arc-cross still depends on
> linux-libc-dev-arc-cross, at least if this was not broken due to related
> changes in cross-toolchain-base-ports.

Sorry you are correct, I thought linux-libc-dev was supposed to provide 
that instead of linux-libc-dev-arc-cross.

> > Therefore I see two options:
> > - linux-libc-dev provides the arc kernel headers
> 
> It does?!?

Yes, correct, sorry about the confusion.

> > - cross-toolchaine-base-ports continues to build 
> >   linux-libc-dev-arc-cross 
> 
> No, src:linux provides linux-libc-dev-arc-cross.  If you think this is
> incorrect, please talk to the CTTE, as this is part of a understanding
> created within their reach last year at DebConf 25.

With a full view of the problem I agree the current solution is fine. 
Therefore I believe the patch you proposed earlier is indeed correct.

Regards
Aurelien