[Reproducible-builds] Cross-building across architectures (reproducibly?)

[Steven Chamberlain]

Steven Chamberlain wrote:
> using Helmut's tool, I've been able to rebootstrap a minimal Debian
> linux-i386 chroot (445 binary packages[3]).  These were cross-compiled
> from source, by only running kfreebsd-amd64 binaries on a FreeBSD
> kernel, and having some Arch:all packages installed.

I've begun to analyze this now, using diffoscope to compare a stage1
debootstrapped chroot of linux-i386, from official Debian packages vs.
my cross-built ones from kfreebsd-amd64.

Unfortunately all the ELF binaries have differences, but most other
files' contents are reproducible.

The biggest problem was that the cross toolchain's linker used a
different target ABI:

│   │   │ -    OS: Linux, ABI: 3.2.0
│   │   │ +    OS: Linux, ABI: 2.6.32

and that leads to huge differences in the output binaries, such as not
using .init_array/.fini_array sections.

That may have consequences, such as if reproducible.debian.net jobs
ever override uname to 2.6.x (Bug #806911), we may run into this problem
there.

umasks are applied to symlinks differently on kfreebsd than on linux:

│   │ -lrwxrwxrwx root/root         0 2015-10-28 22:29:08 ./etc/os-release -> ../usr/lib/os-release
│   │ +lrwxr-xr-x root/root         0 2015-10-28 22:29:08 ./etc/os-release -> ../usr/lib/os-release

base-files seems to look at uname of the build architecture, rather than
host arch it is building for:

│   ├── ./usr/lib/os-release
│   │ @@ -1,6 +1,6 @@
│   │ -PRETTY_NAME="Debian GNU/Linux stretch/sid"
│   │ -NAME="Debian GNU/Linux"
│   │ +PRETTY_NAME="Debian GNU/kFreeBSD stretch/sid"
│   │ +NAME="Debian GNU/kFreeBSD"

gzip embeds the version of a build-dependency, makeinfo.  That's
slightly annoying;  it creates differences where there needn't have
been otherwise:

│   ├── ./usr/share/info/gzip.info.gz
│   │   ├── gzip.info
│   │   │ @@ -1,2827 +1,2827 @@
│   │   │  00000000: 5468 6973 2069 7320 677a 6970 2e69 6e66  This is gzip.inf
│   │   │  00000010: 6f2c 2070 726f 6475 6365 6420 6279 206d  o, produced by m
│   │   │  00000020: 616b 6569 6e66 6f20 7665 7273 696f 6e20  akeinfo version 
│   │   │ -00000030: 352e 3220 6672 6f6d 2067 7a69 702e 7465  5.2 from gzip.te
│   │   │ +00000030: 362e 3020 6672 6f6d 2067 7a69 702e 7465  6.0 from gzip.te
│   │   │  00000040: 7869 2e0a 0a54 6869 7320 6d61 6e75 616c  xi

Then I saw some some reproducibility issues that are already known
(#774425 in dash, #806328 in xz-utils).

Here are my chroots and diffoscope output, in case anyone is curious:

http://kfreebsd.eu/rebootstrap/2016010100/native.tar.xz (12MiB)
http://kfreebsd.eu/rebootstrap/2016010100/rebootstrap.tar.xz (12MiB)
http://kfreebsd.eu/rebootstrap/2016010100/diff.xz (6MiB)

Regards,
-- 
Steven Chamberlain
steven at pyro.eu.org
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: Digital signature
URL: <http://lists.alioth.debian.org/pipermail/reproducible-builds/attachments/20160102/c2fcf375/attachment.sig>