On Tue, Sep 01, 2015 at 12:09:06PM +0200, Dhole wrote:
> As I understand it, the changes in debian/* would be to disable the call
> to po/update.sh. But then there's also changes needed inside po/ because
> the maintainer would need to run update.sh (which would modify files in
> po/) before packaging right? I don't find it clear if the update.sh
> should be run by the Debian maintainer or by upstream.

I consider updating po files to be an "upstream" task.

Sometimes upstream and Debian maintainer are the same person (think
about debconf translations or debian native packages), but even in
such case, updating the files does only need to be done before
dpkg-buildpackage, i.e. only once.

Our task as Debian maintainers is to prepare a source package which is
to be built by the different buildds. It is a little bit odd that
every buildd has to msgmerge the .po files with the updated .pot file.

That's a task for whoever prepares the source package, be it the
Debian maintainer for debconf translations and the like, or the
upstream author for the translations that belong to the upstream
package itself.

> > It's funny that the current gettext fork in alioth did not seem to
> > help about 8 of them:
> > 
> > https://reproducible.debian.net/issues/unstable/lc_messages_randomness_issue.html
> > 
> > (There are 12 unreproducible, but only 8 suffer currently from this issue)
> > 
> The reason why 8 of them are not reproducible even with the patched
> gettext is because the calls to xgettext in those packages don't happen
> under dh, which is one of the current patched toolchains that export
> SOURCE_DATE_EPOCH. Since that variable is not exported, xgettext embeds
> the current date. But that happens to other packages as well with other
> issues (where the toolchain has already been patched to honour

Ok, thanks for the clarification.

So, before telling maintainers "switch to dh and all your LC_MESSAGES
unreproducibility problems will be solved" I'd like they to be aware
that updating po files only once (before dpkg-buildpackage) would be
enough for this to be fixed.

Reproducible-builds mailing list

Reply via email to