Bug#1147469: reproducible-builds: must not set 'nocheck' on only one build

[Christoph Berg]

Re: Thorsten Glaser
> I’ve recently found out that reproducible-builds nowadays fails
> packages which produce different output when 'nocheck' is set vs.
> unset.

This sounds like you have a whole lot of packages with that problem.
Care to share concrete cases?

> This is wrong.

I believe testing for that is very valuable and it should stay.

> If (if!) a 'nocheck' build *profile* is supported by the package,
> then, if that build *profile* is set, the binary packages may not
> differ.

Unless I missed something, packages do not announce profiles they
support. You could only look at debian/rules and guess which profiles
make any difference.

> But the 'nocheck' build option *is* allowed to have binary
> differences, and this is something important for various corner
> cases.

Even if that is true, I think t.rb.o should still test for the
problem. In most cases, the difference should be unintentional and
then people can work on fixing it.

> So, please add it to DEB_BUILD_OPTIONS on either no or all builds,
> and investigate into setting the 'nocheck' build *profile* on only
> one build *iff* the package supports it. I’m not exactly sure how
> to actually test for that, but if it only gets added to the variable
> DEB_BUILD_PROFILES but not DEB_BUILD_OPTIONS, then it can just be
> ignored by packages not supporting the 'nocheck' *profile*.

So the fix is to just set DEB_BUILD_PROFILES=nocheck along and then
the complaint is gone? :)

Christoph