[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: binutils: support toolchain dependency cross translation (aka binutils-for-host/binutils-for-build)

Hi Matthias,

On Sat, Oct 29, 2016 at 11:06:59AM +0200, Helmut Grohne wrote:
> as promised, I was working on the toolchain dependency cross translation
> problem[1]. I believe that my implementation is now in a submittable
> state. The big goal is to introduce packages gcc-for-host and
> gcc-for-build to be used in Build-Depends of other packages to
> disambiguate cross from native compilers. This has to be done in three
> steps gcc-defaults, gcc-N and binutils and we need to start with the
> bottom of the stack (aka binutils). Thus this bug is only concerned with
> binutils.

Thank you for providing a review via irc. I am attaching an updated
patch stack that should addresses all of your remarks.

You requested that binutils-common be Architecture: any instead of
Architecture: all. When the speed of a buildd differs significantly from
the speed of an arch:all buildd, binutils may become temporarily
uninstallable due to an Arch:any -> Arch:all dependency, because dak is
known to throw away outdated packages too early at times. That would
render build-essential uninstallable. Thus I made binutils-common an
Architecture: any data package that is still marked Multi-Arch: foreign.

You remarked that discarding shlibs files results in lintian warnings.
Thank you for highlighting this issue. In this iteration, I have added
the necessary overrides to suppress them.

You asked whether binutils would still cross build after applying my
patch stack. At the time of the original submission binutils would
neither cross build with nor without my patch stack due to #840080.
That bug is now fixed. I verified that binutils builds successfully for
each partial patch application (zero patches to eight patches) in the
following configurations (all builds performed with
 * native build
 * cross building binutils to armel
 * cross binutils targeting m68k and host amd64
 * static cross binutils targeting m68k and host amd64
 * cross binutils targeting hppa64 and host amd64
I had no build failures and the package contents looked reasonable.

It seems that my initial submission of this patch set didn't make it to
d-cross@l.d.o despite X-Debbugs-Cc, so I also hope for some more review
from those readers.

I understand that this patch stack is too late for stretch. Could you
maybe push it (or part of it) to experimental nonetheless?


Attachment: binutils_patches.tar.gz
Description: application/gzip

Reply to: