Rutherther <[email protected]> writes:

> I was considering including the instructions, but didn't want to make
> the e-mail much longer.

Yes, on second thought, leave it as is.  The instructions in the manual
are enough.  It would be more prominent to include reproducibility
instructions in release communications, but, because *at least*
disarchive, guile-git and guile-lib cannot be reproduced, instructions
yield few tiny reproducibility failures in compiled .go files anyway
currently.  Without substitutes, maybe disarchive, guile-git and
guile-lib are the only problem, or maybe more.

Indeed,
SUPPORTED_SYSTEMS=x86_64-linux guix time-machine -q --branch=version-1.5.0 -- 
build -m ./etc/teams/release/artifacts-manifest.scm --no-grafts --derivations

gives

/gnu/store/7rmppjsn47klgdyv2l35izyzg0mx15hj-guix-system-vm-image-1.5.0rc1.x86_64-linux.qcow2.drv
/gnu/store/lc8k57zxqw32bbvigjx2dnk4mm9hr35n-guix-system-install-1.5.0rc1.x86_64-linux.iso.drv
/gnu/store/0lknafwy48l0ijhmxp25j3bc9bdzi6xv-guix-binary-1.5.0rc1.x86_64-linux.tar.xz.drv

and

guix build 
/gnu/store/0lknafwy48l0ijhmxp25j3bc9bdzi6xv-guix-binary-1.5.0rc1.x86_64-linux.tar.xz.drv

downloads a substitute from
https://ci.guix.gnu.org/nar/6kvdzwk4b035nq3vnpp85dvnlq612dcv-guix-binary-1.5.0rc1.x86_64-linux.tar.xz
that gives the same SHA256 hash as in your e-mail, naturally.

> As part of the release, we've moved to generating the artifacts through
> Guile rather than couple of shell lines, so that they can be built more
> easily by CI.

The instructions in the manual are good, too.

diffoscope showed only differences in the tar archive’s disarchive,
guile-git and guile-lib, so if these had been reproducible, the Guile
scripts surely build the same result as the manual’s shell command, just
from a different derivation.  And the manual’s shell command has no
substitutes.

All is good.  Except packages need to be made reproducible someday.

> But in the end, the resulting tarball should be pretty much the same as
> if you did make like you've tried to do.

It is pretty much the same.


> When trying reproduciblity with substitutes, it might be better to
> use --substitute-urls=https://bordeuax.guix.gnu.org in the first
> place.

I guess reproduciblity with substitutes is not proper, anyway.

Thank you for your thoughts!

Regards,
Florian

Reply via email to